ULTRAVIOLET-SELECTED FIELD AND PRE-MAIN-SEQUENCE STARS TOWARD TAURUS AND UPPER SCORPIUS

Modest exposures with GALEX are sensitive to stars as late as G- or K-type at 140 pc, depending on the degree of ultraviolet excess and the degree of extinction. To quantify this limit, we found GALEX 3σ magnitude limits (Table 1) using the GALEX Exposure Time Calculator¹ for exposures of 300 s, 1000 s, and 1500 s, which cover the range of exposure times used by our observations presented in Section 2.2. For comparison, we also give the sensitivity for the GALEX All Sky Survey (AIS) exposure time of 100 s.

In Section 3.1, we use the Two Micron All Sky Survey (2MASS; Cutri et al. 2003) to characterize our GALEX sources, introducing a second set of sensitivity limits. The Explanatory Supplement to the 2MASS All Sky Data Release (Cutri et al. 2008) gives completeness limits both for all detected point sources and for a subset with a photometric quality of A. Both limits are given in Table 2. Since we require only C quality photometry from our 2MASS sources, we expect the completeness limit of our sample to be somewhere between the two values in Table 2. By comparing the location of our fields to Figures 7, 9, 22, and 24 from Cutri et al. (2008, section VI.7.a), we find these limits are unaffected by our fields' proximity to the Galactic plane.

We compare the limiting magnitudes of the GALEX and 2MASS data to the main-sequence magnitudes found in Appendix A. Assuming a distance to both Taurus and Upper Scorpius of 140 pc, a 300 s GALEX exposure can detect unextinguished photospheres down to spectral type F8 in the FUV and K5 in the NUV. However, both Taurus and Upper Scorpius have significant extinction. Assuming A_FUV/A_V = 2.68 and A_NUV/A_V = 2.63 for R_V = 3.1 (A. Gil de Paz 2007, private communication), we can detect stellar photospheres as long as they have an extinction less than that given in the first two columns of Table 3. The 2MASS data are sensitive down to spectral type M, even with extinction, so they do not affect our overall sensitivity.

As we show in Section 3.2, our procedure cannot reliably identify excesses below ∼5.3 mag in the FUV and ∼2.6 mag in the NUV. Therefore, the sensitivity of GALEX does not begin to affect our results until stars with this minimum excess cease to be detectable. The extinctions at which this occurs are given in the last two columns of Table 3. A_V in our Taurus fields is typically less than 3 (Cambrésy 1999), with the exception of the field TauAur_MOS23, where the L1529 dark cloud has extinctions ranging from 3 to 6. A_V in our Upper Scorpius fields is typically less than 2 (Preibisch & Mamajek 2008). Therefore, our observations of stars of type K and later are limited by GALEX sensitivity. Under 2 mag of extinction, GALEX can detect only those K4 stars with an NUV excess of at least 3.7 mag, those M0 stars with an excess of at least 6.9 mag, and those M2 stars with an excess of at least 8.7 mag.

To estimate the excesses we expect from young stars, we take from Valenti et al. (2000) narrow-band continuum fluxes of T Tauri stars measured by IUE at 1760 Å and 1958 Å and assume the stars' FUV and NUV magnitudes correspond to the 1760 Å and 1958 Å fluxes, respectively. The magnitudes are underestimates because broadband photometry will pick up additional flux from emission lines. We estimate that those stars that Valenti et al. (2000) designate as accreting classical T Tauri stars (CTTSs) typically have FUV excesses of 10–18 mag and NUV excesses of 2–7 mag. These stars should be detectable in either the FUV or the NUV down to spectral type K0 under ∼2 mag of extinction. Stars marked as non-accreting naked T Tauri stars can have excesses of up to 10 mag in the FUV and up to 2 mag in the NUV, although some show no continuum excess at all. A 2 mag NUV excess is too small for us to easily distinguish, but we can detect sources with an FUV excess of 5–10 mag down to spectral type F8-G8.

As part of our Cycle 1 proposal, GALEX imaged 61 fields, covering 56 deg², in southern Taurus in the FUV and NUV bands and 13 fields, covering 12.2 deg², in northern Upper Scorpius in only the NUV band. The fields were chosen to coincide with the survey area of Slesnick et al. (2006, 2008). The observed fields are shown in Figures 1 and 2. Because of concerns that bright stars or dense groups of stars might damage the detectors, GALEX avoids fields containing either; unfortunately, these are exactly the fields containing most of the known young stars in these regions. In particular, the observed Taurus fields are located south of the molecular cloud L1527, but north of L1551. We catch the eastern edge of L1529 but avoid L1536. The observed Upper Scorpius fields are located north of all major concentrations of known sources. Most of the pointings were taken in sets of five 300 s exposures of adjacent fields, but a few observations were instead scheduled as full-orbit exposures of 1100–1500 s. The observations are summarized in Table 4.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The observations were reduced automatically by the GALEX data release 2 and 3 pipeline (Morrissey et al. 2007) except for two orbits, TauAur_CH45 and TauAur_CH56, which were processed by an earlier pipeline version. We visually inspected the images and discarded those that were made unusable by clear artifacts or corrupted data.

Two fields, TauAur_CH11_0002_sv03 and TauAur_MOS23, had offsets in their astrometry solutions but no artifacts or signs of data corruption. We found mean astrometry corrections by comparing the positions of eight bright NUV and 2MASS sources in each field, assuming a fixed offset independent of position on the detector. We found that we needed to correct the source positions in TauAur_CH11_0002_sv03 by Δ(α, δ) = (−11.54 ± 026, + 14.97 ± 021) and those in TauAur_MOS23 by (−12.22 ±  023, − 10.73 ±  032) for the GALEX and 2MASS positions to agree. Residuals from the fixed offset (∼08) were larger than the scatter in GALEX–2MASS offsets in comparison fields (∼05) but showed no pattern. The latter is similar to the absolute astrometric precision of 049 found by Morrissey et al. (2007). We applied the corrections to the positions of all GALEX sources in these two fields before considering cross-matches to other catalogs.

We adopted photometry from the band-merged SExtractor (Bertin & Arnouts 1996) catalog produced by the GALEX pipeline. Because GALEX observations have low backgrounds, the pipeline used a customized background algorithm (Morrissey et al. 2007) rather than the one provided with SExtractor. The background was calculated as means over bins of 128 pixels, iteratively clipping pixels having a probability of being Poisson fluctuations below 1.35 × 10⁻³. Bright sources were masked out using an initial SExtractor run. The background map was then linearly interpolated back up to the resolution of the image. The detection threshold map was calculated from the background map as the count rate above which a pixel had less than a 31.7% probability of being a Poisson fluctuation. Except where deblending was needed, SExtractor defined a source as a group of at least 10 adjacent pixels above the local detection threshold.

SExtractor found source fluxes by summing over an elliptical aperture with radius R = 2.5(∑rI(r))/(∑I(r)), which should enclose ∼94% of a source's light for both stars and galaxies (Bertin & Arnouts 1996). Errors were calculated as $\sqrt{A \sigma ^2 + F/g}$ where A is the source area, σ is the background rms, F is the source flux, and g is the gain.² We found that the catalog fluxes are consistent with wide (tens of arcseconds) aperture photometry performed on a trial field. We investigated but did not apply additional nonlinearity or color corrections to the catalog output, as such corrections are still highly uncertain (T. K. Wyder 2008, private communication).

For the remainder of our work, we selected only GALEX sources that: (1) have 3σ flux measurements in at least one GALEX band; (2) SExtractor determined to be a point source with 0.9 or higher probability in each detected band; (3) are within 0.55° of the field center, where PSF variations are negligible, astrometric precision is uniformly high, detector efficiency is high, and artifacts are rare; (4) do not have an artifact flag in the GALEX catalog; and (5) do not suffer any ambiguity in matching FUV and NUV sources.

Twenty-five thousand three hundred and eighty-nine ultraviolet sources in Taurus and 5863 in Upper Scorpius met these criteria. Over 99% of these are unique detections; the rest are multiple detections of sources located in areas of overlap between two GALEX fields.

To characterize our sources, we matched them to data from the 2MASS (Cutri et al. 2003). We chose 2MASS photometry because it is a high quality, well-characterized data set, is available at low Galactic latitudes, and is less sensitive to extinction than visible photometry. We selected 2MASS sources that: (1) have C quality or better photometry in both the J and K bands, i.e., repeatable detections, 5σ flux measurements, and a source profile well fit by a single PSF; (2) are not blended or contaminated in J or K band; (3) are not in the 2MASS extended source catalog; (4) are not associated with a known solar system object; and (5) are within 2'' of a GALEX source, the matching radius recommended by Morrissey et al. (2007).

Twenty thousand and forty-five of the 25389 Taurus GALEX sources and 4210 of the 5863 Upper Scorpius sources had 2MASS counterparts, after removing 184 duplicate GALEX observations from the Taurus sample. There were no duplicates in the Upper Scorpius sample because there were very few overlapping GALEX fields. Figure 3 shows the GALEX–2MASS match fraction as a function of FUV − NUV color, which is relatively insensitive to extinction: E(FUV − NUV)/A_V =  0.05 for R_V =  3.1 (A. Gil de Paz 2007, private communication). Many of the GALEX sources with no 2MASS counterpart have FUV − NUV ≲ 1, the colors expected of star-forming galaxies or B or early A stars. Both types of objects are often too distant to be detected by 2MASS. Even among the redder GALEX sources ∼5% lack 2MASS counterparts. These may simply be sources that were rejected by our quality control criteria.

Zoom In Zoom Out Reset image size

After the merged sources were placed on UV − J versus J − K color–color diagrams (Section 3.2), we found that the source distribution changes rapidly near J ∼  14. Brighter sources tend to form a clear main sequence, as shown in Figure 4, while fainter sources form a different distribution at constant NUV − J ∼ 5. Since we already expected most sources with J > 14 to be unresolved galaxies (Kochanek et al. 2001), we imposed an additional constraint J ⩽ 14 on all our sources.

Zoom In Zoom Out Reset image size

This left 14,130 matched sources in Taurus and 2828 in Upper Scorpius.

Colors calculated by Laget³ from Kurucz models suggest that stellar photospheres form a well-defined locus in the GALEX FUV − J versus J − K and NUV − J versus J − K color spaces, largely independent of surface gravity (Figure 4). Ultraviolet excess sources should appear below this locus, with bluer UV colors relative to non-excess stars of the same photospheric infrared color. Since atmosphere models are not well tested in the ultraviolet, we did not rely on them to determine the precise position of the stellar locus. Instead, we used the population of field stars, a mixture of dwarfs and giants, to define an empirical UV − J versus J − K locus.

Since they can be detected at distances much farther than 140 pc, most of the early-type, intrinsically blue, stars in the field population are background objects, and are therefore extinguished by Taurus or Upper Scorpius. On the other hand, most of the detectable late-type, intrinsically red, stars are in the foreground and suffer little extinction (see Table 3). In practice, since we cannot detect redder stars unless they have an ultraviolet excess, our measurement of the field star locus is determined by early-type stars, which have a similar extinction to those in Taurus or Upper Scorpius (A_V ∼ 2). In addition, the reddening vector is almost parallel to the main sequence, and as a result the locus of reddened stars inferred, e.g., from the Kurucz model colors is much narrower than the stellar locus we observe. Since we are interested in the distance of a star from the locus, we infer that any systematic errors associated with reddening are much smaller than the effect of other errors or astrophysical variations.

Although we tried to characterize the main sequence by taking the median NUV − J color of narrow J − K bins, following Covey et al. (2007), this approach failed for bins with J − K ≳ 0.5 in the FUV and J − K ≳ 0.9 in the NUV, as there were too few field stars to give an unbiased median. Instead, we iteratively removed points that differed from the median locus by three standard deviations, then made linear fits to the remaining points with J − K ⩽ 0.4 in the FUV and J − K ⩽ 0.8 in the NUV. The resulting solutions were

$$\begin{eqnarray} (\mbox{FUV}-J) &=& (20.5 \pm 0.5) (J-K) + (2.96 \pm 0.14)\nonumber \\[3pt] \hbox{cov}(m,b) &=& -0.07 \quad \hbox{(Taurus)}\nonumber \\[3pt] (\mbox{NUV}-J) & = & (10.36 \pm 0.07) (J-K) + (2.76 \pm 0.04)\nonumber \\[3pt] \hbox{cov}(m,b) &=& -2\times 10^{-3} \quad \hbox{(Taurus)}\nonumber \\[3pt] (\mbox{NUV}-J) & = & (10.05 \pm 0.16) (J-K) + (2.64 \pm 0.08)\nonumber \\[3pt] \hbox{cov}(m,b) &=& -0.013 \quad \hbox{(Upper Scorpius)}. \end{eqnarray} \tag{ 2 }$$

We defined a source's FUV and NUV excess as the difference between its observed FUV − J and NUV − J color, respectively, and the value predicted by Equation (2). The formal error in the excess, propagated from the photometric uncertainties and the uncertainties in the fit, is dominated by the error in the J − K color. Because the stellar locus has a steep slope in (UV − J)–(J − K) space, the mean uncertainty of 0.05 in J − K propagates to an uncertainty of ∼1.1 mag in the FUV excess and ∼0.5 mag in the NUV excess. Since in Section 3.3, we identify our sample of UV excess sources as those differing by a certain number of standard deviations from the main stellar locus, this translates to the smallest excess we can reliably detect: 3.2 FUV magnitudes and 1.6 NUV magnitudes, if we select sources with a 3σ or larger excess; or 5.3 FUV magnitudes and 2.6 NUV magnitudes, if we require a 5σ or larger excess. For stars with more precise J − K colors, we can detect excesses below these limits.

We show in Figure 5 the distribution of excesses divided by their errors. The observed distributions are roughly 1.9 times broader in the FUV than the errors imply, and 1.5 times broader in the NUV. This might suggest a source of error we have overlooked. An alternative is that the locus of field stars is broadened by variations in properties other than temperature and gravity, and we are seeing a convolution of such astrophysical broadening with our errors. We will continue to quote excesses in multiples of the formal errors as a measure of whether a source lies outside the locus, but the reader should bear in mind that this ratio is not a rigorous measure of statistical significance. However, in sparsely populated areas of the color–color diagram, especially at high J − K, there are too few sources for us to use alternative measures such as the locus width.

Zoom In Zoom Out Reset image size

We now define our sample of UV excess sources as those having ultraviolet excess normalized by the formal error in that excess above a given cutoff value. For several choices of cutoff (3σ, 4σ, ... , 7σ), we estimate the number of sources erroneously included in the UV excess sample by assuming that the distribution of field star colors is symmetric with respect to positive and negative excesses. For example, to estimate the number of erroneously included stars whose measured UV excess is 3σ or greater, we find the number of sources with a 3σ or greater UV deficit and subtract it from the number of UV excess sources, assuming the former are all outliers from the field locus. We list in Table 5 the total number of UV excess sources, the expected number of erroneous sources, and the fraction of UV excess sources that appear legitimate for each cutoff value.

Table 5 shows that the choice of cutoff carries a steep trade-off between completeness of the UV excess sample and reliability of an individual source's identification as UV excess. For example, only a third of the sources with a measured excess of over 3σ should be true UV excess sources, while around 86% of the 5σ sources are reliable identifications. On the other hand, only 42% of the genuine 3σ UV excess sources are expected to appear in a 5σ sample—the rest are buried in the more heavily contaminated population with excess between 3σ and 5σ. To support projects where completeness is a higher priority than reliability, we present in Table 6 all sources with a 3σ or better excess in at least one band. However, for many of our discussions in this paper, we will concentrate on the more reliable 5σ subsample.

Table 6. Sources with a 3σ or Greater UV Excess in at Least One Band

	α	δ		FUV	σFUV	NUV	σNUV	ΔFUV	$\sigma _{\Delta _{\rm FUV}}$		ΔNUV	$\sigma _{\Delta _{\rm NUV}}$
2MASS	(deg)	(deg)	GALEX Tile	(mag)	(mag)	(mag)	(mag)	(mag)	(mag)	ΔFUV/σ	(mag)	(mag)	ΔNUV/σ
03171986+2447143	49.332784	24.787331	TauAur_MOS44_v2	⋅⋅⋅	⋅⋅⋅	20.187	0.066	⋅⋅⋅	⋅⋅⋅	⋅⋅⋅	1.5	0.5	3.1
03174690+2429233	49.445453	24.489820	TauAur_MOS44_v2	⋅⋅⋅	⋅⋅⋅	20.273	0.068	⋅⋅⋅	⋅⋅⋅	⋅⋅⋅	2.1	0.4	5.0
03180156+2411297	49.506523	24.191601	TauAur_MOS44_v2	⋅⋅⋅	⋅⋅⋅	20.085	0.061	⋅⋅⋅	⋅⋅⋅	⋅⋅⋅	1.7	0.4	4.1
03181321+2447174	49.555079	24.788185	TauAur_MOS44_v2	⋅⋅⋅	⋅⋅⋅	18.324	0.021	⋅⋅⋅	⋅⋅⋅	⋅⋅⋅	1.3	0.3	4.2
03181487+2429114	49.561995	24.486513	TauAur_MOS44_v2	⋅⋅⋅	⋅⋅⋅	19.905	0.052	⋅⋅⋅	⋅⋅⋅	⋅⋅⋅	1.8	0.4	4.5

Notes. Positions are in J2000.0 and taken from the 2MASS Point Source Catalog. FUV and NUV photometry in AB magnitudes is taken from the GALEX SExtractor catalogs. J, H, and K photometry in Vega-based magnitudes is taken from 2MASS. The notation Δ_FUV denotes the FUV excess. If we know of an association between the GALEX source and a known object, and if the literature confirms or rules out membership in Taurus or Upper Scorpius, we list the information in the last two columns. Our list of associations is not complete.

Only a portion of this table is shown here to demonstrate its form and content. Machine-readable and Virtual Observatory (VO) versions of the full table are available.

Download table as:  Machine-readable (MRT)Virtual Observatory (VOT)Typeset image

The spatial distribution of the 5σ excess sources is shown in Figures 6 and 7. Most of our sample of 5σ sources in both Taurus and Upper Scorpius is spread uniformly over the observed fields.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

In Taurus, we also find concentrations of sources near α = 80°, δ = 24° and near α = 62°, δ = 25°, as well as some outlying members of the previously known group of sources associated with the cloud L1529 (Kenyon et al. 2008). The first group might not be a true physical association, appearing dense only because it is less than 10° from the Galactic plane. The group near (62°, 25°), consisting of six sources, is likely a Poisson fluctuation. Given the mean density of 1.5 5σ sources per field, we find a 21% probability that at least one of our 61 Taurus fields contains 6 sources or more. We conclude there is no statistical evidence for clustering of the UV excess sources.

There are very few sources in our Upper Scorpius sample, consistent with the observation by Slesnick et al. (2008) that only a small fraction of Upper Scorpius members of spectral types M3-M8 are north of δ = −17°. The sample is too sparse for us to look for evidence of structure, nor has any previously been observed in the association (Preibisch & Mamajek 2008).

We examine in Section 6.1 whether any of these UV excess sources are plausible members of Taurus or Upper Scorpius.

To test our survey's recovery rate, we identified samples of Taurus and Upper Scorpius members compiled without reference to our observations. Kraus & Hillenbrand (2007a) and Walter et al. (1988) together list 261 known Taurus members and Kraus & Hillenbrand (2007a) list in addition 401 Upper Scorpius members. Unfortunately, because of concerns that a bright UV source or concentration of sources might damage the GALEX detectors, our GALEX pointings avoid areas with large numbers of known young stars (Figures 1 and 2). Only five of the previously known Taurus members were in the GALEX survey region, and none of the known Upper Scorpius members; the former are listed in Table 7.

We recover AA Tau and DN Tau, both well-known classical T Tauri stars, as UV excess objects to high significance. TAP 4, a naked T Tauri star, has only a 3σ UV excess. The remaining two sources, which we do not detect, are M stars, and therefore strongly selected against by GALEX without strong UV excesses (Section 2.1). While the detections of AA Tau and DN Tau are reassuring, the marginal detection of TAP 4 suggests that many of the more mature Taurus members—our main goal—may be buried in the UV-bright tail of the field star population, possibly below our 3σ limit.

To investigate the amount of field contamination in our 3σ UV excess sample, we repeated our procedure on a set of seven Medium Imaging Survey (∼1400 s exposure) fields, covering 6.5 deg², taken near (l = 90, b = 30). This latitude is slightly higher than that of Taurus or Upper Scorpius, but there were no GALEX observations at lower latitudes. The longitude was chosen as a point intermediate between the bulge (the direction of Upper Scorpius) and the anticenter (the direction of Taurus). IRAS images of this region show little emission aside from Galactic cirrus, and there are no known open clusters within 10° of the fields.⁴ Based on these observations, we do not expect a population analogous to the Taurus or Upper Scorpius association in our comparison fields.

Of the 10,496 GALEX sources that met our quality cuts, 3148 had 2MASS matches. From examining the source distribution in color–color space, we confirmed that J ⩽ 14 discriminated stars from unresolved galaxies in these fields, but only 1461 sources met this last criterion. The small fraction of matching sources, and stellar sources in particular, is expected from the higher galactic latitude and lower extinction of our control fields compared to our target fields. For reference, 14,130 of our 25,389 Taurus GALEX sources had a 2MASS counterpart with J ⩽ 14, as did 2828 of our 5863 Upper Scorpius GALEX sources. Because of the low extinction, our sample of GALEX sources in the control fields contains on average more distant sources than the samples in the target fields do. We show in Figure 8 the distribution of sources as a function of FUV − NUV color, analogous to Figure 3 and the discussion in Section 3.1. Very few of the sources with FUV − NUV ≲ 2 (either stars earlier than mid-A or unresolved star-forming galaxies) have 2MASS counterparts; the effect is much stronger here than in Figure 3. In addition, the distribution of J magnitudes for matched sources (Figure 9) peaks over a magnitude fainter than in our target fields, suggesting again that we are probing greater distances in the control field.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We show in Figure 10 color–color diagrams for the control fields analogous to Figure 4. These fields contain nine 5σ UV excess sources. This is the same fraction of GALEX–2MASS sources (0.6% ±  0.2%) as in Taurus (0.63% ±  0.07%) and Upper Scorpius (0.42% ± 0.12%). We cannot tell what fraction of the excess sources in the control fields are outliers from the main stellar locus, as there are no UV-deficit sources in Table 5 with which to repeat the analysis of Section 3.3. Since these sources appear far from known clusters or star-forming regions, we assume they represent a population spread evenly throughout the Galactic disk. If this is the case, then the population should appear along any line of sight at similar Galactic latitude, including toward Taurus and Upper Scorpius, and in fact should account for a large fraction of the sources we see there.

Zoom In Zoom Out Reset image size

This inferred population of UV excess sources is qualitatively similar to the population of ∼100 Myr old G and K stars invoked by Briceño et al. (1997) to explain ROSAT detection rates of X-ray luminous stars. However, they predicted a surface density of 0.2–0.3 deg⁻² for these stars, implying an average of 1.3–1.9 such stars in an area the size of our control fields. Because Briceño et al. (1997)'s work was based on ROSAT's sensitivity limits, most of the detectable stars in their model are within 100–200 pc. Three of the nine UV excess stars in the control sample have J magnitudes too faint for a distance of 200 pc, so we drop these three to allow a fair comparison of our sample, now reduced to six, to their results. The probability that we would observe six or more sources when we expect 1.9, however, is only 1.3%. Since to our knowledge there is no study on low-latitude ultraviolet populations analogous to Briceño et al. (1997), we cannot explain why our background of UV excess sources is much larger than their background of X-ray sources.

Like all youth indicators, ultraviolet excess is not perfectly correlated with age (F. Altenbach 2010, in preparation). As a result, like most young star surveys, we need spectroscopic follow-up to confirm the youth of our photometrically selected stars. We carried out a spectroscopic observing program of some of our 3σ ultraviolet excess sources using the Double Spectrograph (DBSP) on the 200 inch Hale Telescope at Palomar. Our program had several goals. One was to obtain representative spectra of excess sources across a broad area of color–color and color–magnitude space. Another was to understand the diversity of sources that produce a UV excess. A third was to create as large a sample of spectroscopic candidate young stars as possible.

To balance these goals, we created a spectroscopic target list containing the following groups of sources, in decreasing order of population: (1) UV excess sources with J − K > 0.7 and 10 < J < 14, as we found from our first run that this region of parameter space contained many emission-line stars; (2) UV excess sources with 0.3 < J − K < 0.5 and 10 < J < 12, as this region of color–magnitude space was dominated by sources with an excess only in the FUV, suggesting they represented a different population from sources in other regions; (3) UV excess sources with J − K > 0.5 and an FUV excess, as these sources tended to have very large excesses with no clear pattern; and (4) obvious outliers in Figure 4, either sources with J − K much lower or much higher than the main body of UV excess sources or sources with much larger excesses than other UV excess sources with similar J − K.

At the telescope, we would usually choose the brightest sources consistent with the above criteria to keep the exposure times reasonable. In addition, we tended to pick sources with large excesses (≳7 mag in FUV or ≳3 mag in NUV) as a previous program of AIS-selected observations had shown that these sources are much more likely to have chromospheric activity indicators in their spectra (F. Altenbach 2010, in preparation). As a result, the spectroscopic sample has strong biases not present in the photometric sample of Section 3.4.

We observed 20 of the 95 3σ candidates in Upper Scorpius, including 7 of the 11 5σ candidates, on 2008 June 5–6. We also observed Sco X-1, but do not count it as a candidate as we were aware of its identity before the run (see Section 6.2). We likewise observed 43 of the 471 3σ candidates in Taurus, including 20 of the 89 5σ candidates, on 2008 November 28. The spectra covered the ranges 3890–5430 Å at signal-to-noise ratio (S/N) ∼ 50–110, R ∼ 6400 and 6190–6750 Å at S/N ∼85–160, R ∼ 8600, allowing us to observe the key youth indicators Ca ii emission (3934, 3968 Å), Hβ emission (4861 Å), Hα emission (6563 Å), and Li absorption (6707 Å). The images were reduced using the JHU astronomy library in IDL, and the spectra were extracted with the NOAO two-dimensional spectral package in IRAF.

In Table 8, we list all the Palomar targets, together with equivalent widths of the youth indicators where we could measure them. In addition, where we did not detect lithium we give an upper limit on the equivalent width of 10 Å/S/N, where S/N is the formal S/N of the surrounding continuum, as propagated by the IRAF spectral extraction routines. The constant of proportionality was set by noting that, had the lithium line in 2MASS J04505356+2139233 been weaker than its 0.15 Å value, we would have taken it for a 3σ statistical fluctuation.

The spectra were classified visually using spectra of standards taken on a previous DBSP run, with the statistics shown in Table 9. As expected, our Upper Scorpius sample is dominated by G and K stars, the lowest mass (and therefore most numerous) stars to which we are sensitive. On the other hand, our Taurus sample shows a large number of M stars but very few K stars, particularly with a 5σ or greater excess. This is surprising because while K stars with weak excesses are not detectable by GALEX, those with a 3.5 mag (≳7σ) excess or low extinction should be (Section 2.1). Table 8 has many stars of other spectral types, including later types, with excesses of this level. We suspect the deficit of K stars in our sample may represent an unexpected selection effect in our choice of spectroscopic targets.

We found that roughly a third of our sources had chromospheric activity indicators (Table 9). Three of those, all in Taurus, also show lithium absorption. We present spectra of these three in Figure 11. Among the three lithium stars, our survey recovers one previously known Taurus member, 1RXS J044712.8+203809, and confirms a previous candidate, 1RXS J045053.5+213927. The third star, 2MASS J05122759+2253492, is a previously unknown lithium-rich M dwarf toward the Galactic plane. The self-absorbed Hα profile suggests this star is still accreting, and the high Li equivalent width of 530 mÅ implies an age under 10 Myr.

Zoom In Zoom Out Reset image size

While many of our emission-line stars in Taurus show both Balmer and calcium emission lines, all but one in Upper Scorpius show only calcium emission. Since, unlike Balmer emission, calcium emission is a chromospheric activity indicator often seen in field stars, this may mean that our Upper Scorpius fields are too far from the main body of the association to contain significant numbers of young stars and that nearly all of our Upper Scorpius candidates are ≳100 Myr old field stars.

Two thirds of the spectra, even of 5σ targets, show no unusual features, only photospheric absorption lines. This is surprising because we anticipated a good correlation between UV photometric excess and the presence of optical emission lines, aside from the population included from statistical fluctuations that we estimated in Section 3.3. From Table 5, we expect only one sixth of the 5σ targets to be erroneously identified as UV excess, and therefore for five sixths, not the observed one third, of the targets to show emission lines. We conclude there is a significant population of UV excess sources with normal optical spectra; we explore this population more in Section 6.3.

With our spectroscopic data, we can investigate how much of our UV excess sample, most of which lacks spectra, can be associated with Taurus or Upper Scorpius. A significant number of our spectra are of sources found below the main sequence at 140 pc, in part because we wanted to characterize our contaminants, in part because we were biased toward optically bright (and therefore blue, at fixed J) sources, and in part because we were concerned that our brightest sources were too bright even for pre-main-sequence stars and must instead be field giants. Of the 19 3σ sources above the main sequence in Upper Scorpius, we obtained spectra of 9. Of the 101 3σ sources above the main sequence, we followed up 23. In addition, we found information in the literature confirming or ruling out Taurus membership for 14 sources, two of which were also spectroscopic targets. We searched for but found no such information for our Upper Scorpius candidates. We summarize these results in Table 10, classifying UV excess stars above the 140 pc main sequence as follows: (1) stars confirmed as members of Taurus or Upper Scorpius in previous literature, usually by lithium detection; (2) stars rejected as members in previous literature, usually by lithium non-detection or by proper motion association with the nearby Pleiades or Hyades; (3) Palomar targets with no emission lines; (4) Palomar targets with emission lines, but with a spectral type or luminosity class indicating they are actually either reddened background main-sequence stars or background giants; (5) Palomar targets with spectral type consistent with Taurus or Upper Scorpius membership and emission lines but no lithium line; (6) Palomar targets with spectral type consistent with Taurus or Upper Scorpius membership and emission lines and a lithium detection; and (7) stars with neither membership information in previous literature nor Palomar spectra.

For the five stars in Taurus and five in Upper Scorpius that had both emission lines and a spectral type consistent with Taurus or Upper Scorpius membership, we acquired preliminary proper motions based on published USNO and 2MASS positions (A. Kraus 2009, private communication). The results, presented in Table 11, indicate that three of the spectroscopic candidates in Taurus are likely members of the association, while only one of the candidates in Upper Scorpius is. However, we emphasize that these are very rough proper motions and should be used with caution.

We show in Figure 12 the spatial distribution of our spectroscopic targets and emission-line stars in Taurus. We show in Figure 13 color–magnitude diagrams of the Taurus, the Upper Scorpius, and the control fields, distinguishing both 5σ UV excess sources and spectroscopic targets (details in caption). We show the main sequence at 140 pc in blue. We expect low-mass Taurus and Upper Scorpius members to lie above the indicated main sequence, as would extinguished early-type main-sequence stars at a comparable distance. Figure 12 shows that, as expected, almost all of our Palomar targets with emission lines are above the 140 pc main sequence, to within uncertainties in the stars' J − K color. The rarity of emission lines among the background UV excess stars suggests that the emission-line stars above the main sequence may represent a population associated with Taurus or Upper Scorpius, although we are limited by small numbers: a Fisher's exact test gives a probability of 20% that we would see at least this large a discrepancy from statistical fluctuations if the fraction of emission stars were spatially uniform.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

In Section 3.4, we found many UV excess sources in the easternmost fields of our Taurus survey, but noted that these are the fields closest to the Galactic plane. To determine the nature of these fields, we plot their sources in Figure 14, with the fields in central Taurus for comparison. Clearly, the vast majority of the 5σ sources in the eastern fields are background objects; the high density of UV excess sources merely reflects the high density of stars toward the Galactic plane.

Zoom In Zoom Out Reset image size

To see if the high density of emission-line stars in eastern Taurus is likewise due to the fields' low Galactic latitude, we repeat a test from Slesnick et al. (2006), who carried out an optical survey of roughly the same area as our GALEX survey. By selecting candidate pre-main-sequence stars from optical color–magnitude diagrams and using follow-up spectroscopy to look for gravity indicators, they found a broadly distributed population of pre-main-sequence stars at α ≳ 80°. They found no such population in western Taurus. Using Fisher's exact test, they showed that the probability of observing so great a difference between the fraction of spectroscopically confirmed pre-main-sequence stars east and west of the known Taurus population is less than 3% if, in fact, the two regions represent identical populations.

Applying Fisher's exact test to our own results (four out of 15 Palomar targets have emission lines east of Taurus and two out of eight have them west of Taurus), we find a probability of 100% of observing fractions of emission-line stars differing by at least as much as they do. Repeating the test on samples restricted to 5σ sources, or samples above the main sequence, likewise gives no significant result. We can explain the insignificance of our result in part as an effect of selecting sources only by their colors, making us more sensitive to background objects than Slesnick et al. (2006), and in part by the small number of spectra we have. If we had a larger sample, we might be able to get a more useful constraint on the existence of an eastern extension to the Taurus association.

Despite the strong backgrounds, we have identified possible new members of Taurus and Upper Scorpius, with more likely to be identified once we have additional spectra. Since our GALEX fields do not cover the known areas of newly formed stars in either Taurus or Upper Scorpius, any new members we find must represent either previously unnoticed subgroups of Taurus (though we find no significant clustering), a more distributed mode of star formation than previously inferred in either region (though this is hard to reconcile with the classical theory that stars form in dense clumps), or migration from the actual sites of star formation (Feigelson 1996; Briceño et al. 1997). In principle, we can constrain this last hypothesis with the toy model we presented in the introduction to this paper. However, as we find only one new candidate of type K2 or earlier in our Taurus survey, in addition to the three already known from the literature, and three new candidates of type K2 or earlier in Upper Scorpius, we do not yet have a sample large enough for good constraints. Our single detection in Taurus implies that 13% of our 3σ sources might be K2 or earlier members. Extrapolating to the 36 3σ sources in our central Taurus fields, where the calculations of the introduction apply, we predict a member density of 0.3 deg⁻². Our three K2 or earlier candidates in Upper Scorpius imply that 33% of our unobserved sources might also be members, implying a member density of 0.5 deg⁻². Assuming that half of the sources in either region turn out to be zero-age main sequence stars or other field contaminants, we find densities similar to the ones we predicted for constant star formation histories of 30 Myr or longer. Trying to apply formal upper limits rather than these highly uncertain values gives no constraint at all.

Since most of our emission-line stars lack lithium detections, our arguments that they truly are Taurus or Upper Scorpius members are statistical in nature. Better constraints on the ages of these stars could tell us whether they are co-eval with the known Taurus and Upper Scorpius populations or whether they are a distinctly older group—since stars of given age show a broad scatter in their lithium equivalent widths, our non-detections are suggestive but do not by themselves prove an old population. In addition, more rigorous proper motions could confirm individual stars' membership and provide cleaner statistics, especially once we have an expanded spectroscopic sample.

While our survey targeted chromospherically active young stars in Taurus and Upper Scorpius, GALEX is an ideal facility for finding a variety of UV-bright sources, potentially including new classes of objects. However, this diversity means that any search for a specific type of object must consider a variety of interlopers. As a demonstration of both the power of UV surveys to find unusual objects and the potential contaminants of such surveys, we note that our photometrically selected sample includes an X-ray binary, a possible mass-transfer binary, and a white dwarf M dwarf (WDMD) pair.

2MASS J16195506 − 1538250 (Sco X-1). At J − K = 0.76, NUV − J = 2.05, Sco X-1 is a clear outlier in Figure 4. We observed it at Palomar so that we would have a template spectrum for other X-ray binaries that might appear at smaller UV excess. Since there are no other sources in our Palomar sample with similar (emission-line dominated) spectra, and since we deliberately selected high-excess sources as Palomar targets, we are confident this is the only X-ray binary in our sample.

2MASS J05161463+2313174. This A0e V star has a self-absorbed Hα emission line and an inverse P Cygni Hβ line (Figure 15), both signs of accretion. The difference between the peak and the trough of the Hβ line is about 200 km s⁻¹. The star's J magnitude of 10.8 and a diffuse interstellar band at 4430 Å are difficult to reconcile with Taurus membership. It may instead be a background (≳600 pc away) mass-transfer binary; however, we note that it has a much more modest UV excess than Sco X-1 (Table 8).

Zoom In Zoom Out Reset image size

2MASS J04015065+2103495. This star's spectrum contains both broad Balmer absorption lines and strong TiO bands, identifying it as a white dwarf M dwarf double. It is found at J − K = 0.83, NUV − J = 6.67 in Figure 4. While it is at the edge of the locus of UV excess sources, it is not an obvious outlier as was Sco X-1. As a result, WDMD pairs are a potentially significant contaminant in our sample.

To identify the locus of WDMD doubles in UV − J versus J − K space, we find empirical absolute magnitudes of white dwarfs in Appendix B. In particular, a white dwarf with T_eff ∼ 16000 K has M_FUV = 11.11, M_NUV = 11.41. A double with an M0 primary and a 16000 K secondary, then, has FUV − J ∼ 5.1, NUV − J ∼ 5.4, J − K ∼ 0.86 (or an apparent FUV excess of 20 mag and NUV excess of 7.3 mag over the M dwarf photosphere). As their white dwarfs cool, WDMD pairs form the WDMD locus by migrating along lines of constant J − K toward the main sequence. Any extinction moves the WDMD locus redward almost parallel to the field locus.

Smolčić et al. (2004) report that one out of every 2300 Sloan Digital Sky Survey (SDSS) stars with u <  20.5 are WDMD pairs, while Seibert et al. (2005) find that using NUV data from the GALEX Early Release rather than the u band doubles this number. Given 14,130 and 2828 matched sources in Taurus and Upper Scorpius, respectively, we expect about 12 and two of them to be WDMD pairs. Since it is not clear to what excess level Seibert et al. (2005) were able to probe, we cannot convert this figure to a number of UV excess sources in our sample; their study, benefiting from the high precision of SDSS photometry, may be sensitive to WDMDs to which we are not.

We identified in Section 6.2 examples of sources other than chromospherically active stars that produce an ultraviolet excess, with spectra characteristic of X-ray binaries or WDMD pairs. However, many other UV excess sources remain unexplained, as their spectra are those of ordinary main-sequence stars. In Taurus, 12 of our 20 5σ targets show no signs of chromospheric activity in their spectra. Extrapolating to the rest of the 5σ sample, we infer a density of UV excess sources without optical activity indicators between 0.7 deg⁻² and 1.2 deg⁻² (90% symmetric confidence interval). In Upper Scorpius, the spectra of four of our seven 5σ targets show no unusual features. We infer a density of UV excess sources without optical activity indicators between 0.4 deg⁻² and 0.7 deg⁻² with 90% confidence. We might have overlooked chromospheric activity indicators among A or early F stars, where the strong photospheric background makes it difficult to discern optical emission lines (e.g., Simon & Landsman 1991), but we should have seen any such indicators among the later spectral types.

We cannot attribute the entire population of stars with UV excesses but normal optical spectra to errors in our photometry or in our characterization of the field locus. Based on the statistics in Table 5, we expect three of our 5σ Taurus spectroscopic targets to be included by such errors, and at most one in our Upper Sco sample. In addition, we see stars with normal spectra at up to 10σ excess, where such uncertainties should have no effect. We must seek other explanations for this result.

The first possibility is that we are looking at chance superpositions of UV and infrared sources. To find the expected number of such sources, we model the GALEX and 2MASS sources in our fields as three uniformly and independently distributed populations: one of UV-only sources with density Σ_UVonly, one of IR-only sources with density Σ_IRonly, and one of sources visible in both bands with density Σ_both. The observed densities of UV sources, IR sources, and matched sources should then be

$$\begin{equation} \begin{array}{r@{\,}c@{\,}l} \Sigma _{\rm UV} & = & \Sigma _{\rm UVonly} + \Sigma _{\rm both} \\ \Sigma _{\rm IR} & = & \Sigma _{\rm IRonly} + \Sigma _{\rm both} \\ \Sigma _{\rm match} & = & \Sigma _{\rm both} + \Sigma _{\rm UVonly}\Sigma _{\rm IRonly}A_{m} \\ \end{array}\!, \end{equation} \tag{ 3 }$$

where A_m = π(2'')² is the area around a GALEX source in which a 2MASS source would be assumed a counterpart. Taking values for Σ_UV, Σ_IR, and Σ_match from Table 4 (note, we only count 2MASS sources with J ⩽  14, as we have throughout the paper) and solving for the density of chance alignments Σ_UVonlyΣ_IRonlyA_m, we expect six such matches in our Taurus survey region and two in our Upper Scorpius fields. Since there are 89 5σ UV-excess sources in Taurus and 11 in Upper Scorpius, and since some chance matches might produce a smaller excess than 5σ, we can discount them as the source of most of these unexplained UV excesses.

J. M. Carpenter (2009, private communication) has suggested, based on Hogg & Turner (1998), that some of our UV excess sources, those detected at low S/N, have overestimated UV fluxes. If this flux bias is responsible for our unexplained UV excess sources, then we expect low S/N GALEX sources to be more likely to show a UV excess. However, when we measure the fraction of UV excess sources as a function of the GALEX S/N, we find no significant dependence on the S/N. S/N-related biases do not explain why so many UV excess sources appear inactive in spectroscopic follow-up.

Bianchi et al. (2007) have suggested that spurious UV excess sources may arise from confused matches in which a single GALEX source has multiple 2MASS counterparts. Our matching radius of 2'' does not let us find more than one of a GALEX source's 2MASS counterparts, so we cannot easily identify sources where this occurs. As a test of whether overlooked counterparts can account for our excess sources, we match our GALEX sample to the All Sky Combined Catalog (Kharchenko 2001), finding four close doubles detected by GALEX: HD 21392, HD 26514, HD 141959, and HD 242903. Of these, HD 26514 was excluded from our 2MASS sample because it had only E quality photometry in J band (profile fit failed or intermittent detection), HD 242903 because it had F quality in K band (error could not be determined), and HD 141959 because it had a confused K detection. Only HD 21392 was included in our 2MASS sample, and it had an NUV − J color consistent with a single photosphere (0.3σ excess). This test suggests that our quality cuts reliably exclude marginally resolved doubles and our procedure is not susceptible to this source of error.

As a second test of the confusion hypothesis, we examine the fraction of UV excess sources without spectroscopic activity indicators as a function of the offset between GALEX and 2MASS positions. If either the GALEX or 2MASS source is a marginally resolved double, we expect its measured position to have a greater offset from its true position than if it were a single source, and therefore a greater offset from an independent measurement in another band. If this is the cause of our spectroscopically unconfirmed UV excess sources, we expect the apparently inactive sources to be biased toward larger GALEX–2MASS separations. In Figure 16, we show a histogram of the Palomar targets as a function of the separation. 32 of the 44 sources (73%) with separations less than 1'' lack optical activity indicators, whereas 11 of the 19 sources (58%) with separations between 1'' and 2'' do. Assuming, as a prior, that the parent fraction of sources with no activity indicators is uniformly and independently distributed over the interval [0, 1] in either group, we find only a 12% probability that the high-separation targets were drawn from a population with a larger such fraction than the low-separation targets.

Zoom In Zoom Out Reset image size

Having considered several forms of experimental error, we now consider what contaminating sources could produce an ultraviolet excess but a normal optical spectrum. One possibility is a binary consisting of a G- or K-type primary and a flare star secondary. In Appendix C, we show that the flare star CR Dra has M_NUV = 14.09 in quiescence and M_NUV = 13.27 if observed during a flare. A K0 star with a companion similar to CR Dra would show an apparent UV excess of 0.15 mag in quiescence and 0.29 mag if we caught it during a flare. Only for a K4 or later primary would a companion flare star produce apparent excesses of the right order (1.7 mag in quiescence, 2.4 in flare) to be detectable by our procedure. Allowing for a flare ∼2 mag brighter makes this a potential source of contamination for K2 and later stars. However, since most of our UV excess sources with normal spectra are around G or earlier stars (cf. Table 9), we cannot explain the NUV excesses as flux from a companion flare star. We do not have any data with which to estimate the effect of a flare star companion on a system's FUV magnitude.

A similar alternative is a binary consisting of a G- or K-type primary and a white dwarf secondary. A G or K star has an NUV flux comparable to that of a white dwarf, so such binaries are most prominent in the FUV. Taking values from the appendices, a K0 star with a 16,000 K white dwarf companion would have an apparent NUV excess of only 1 mag, but a clear apparent FUV excess of 10 mag. A G0 star and a 16,000 K white dwarf would have a somewhat more marginal apparent FUV excess of 5 mag. Since a 0.6M_☉ white dwarf cools to 16,000 K within 150 Myr, and even a 1.0M_☉ dwarf cools within 450 Myr (Fontaine et al. 2001), binaries with white dwarfs hotter than 16,000 K should be too rare to contribute a significant population of NUV excess sources. G and K stars with white dwarfs might explain some of our FUV-only excess sources, but nothing with an NUV excess.

Five of our 20 5σ spectroscopic targets in Taurus and one of our seven 5σ targets in Upper Scorpius are late F- or G-type stars with an NUV excess and normal optical spectra. None of the explanations discussed here are plausible for these stars, and they will form the subject of a future paper once additional follow-up is obtained.