A REVISED AGE FOR UPPER SCORPIUS AND THE STAR FORMATION HISTORY AMONG THE F-TYPE MEMBERS OF THE SCORPIUS–CENTAURUS OB ASSOCIATION

The optical spectra were visually classified using the primary temperature classification spectral features for F-type stars: the strength and profile of the Balmer lines, followed by several metal lines including the G band at ∼4310 Å (Gray & Corbally 2009). PMS stars may exhibit enhanced chromospheric activity, which will weaken the strength of the Balmer absorption lines or they may be seen in emission. In this case a correct classification can still be obtained using the wings of the hydrogen line profiles (R. O. Gray 2010, private communication). In addition to the hydrogen lines and the strength and shape of the G band, we used the strength of the temperature-sensitive Ca i λ4226 and Fe i λ4383 lines to confirm the classifications, although in this temperature region we found them less useful than the G band.

Stellar spectral classification is greatly facilitated by the ability to visually overlay the object on spectral standards obtained with the same spectral resolution. To that end, we developed a visual classification tool (sptool) which presents the object spectrum against two standard star spectra and allows the user to change the comparison standards with a single keystroke. The tool allows the classifier to easily move in temperature and luminosity subtype space by using the arrow keys on the keyboard. The tool, written in Python, is freely available online⁵ and only requires a dense grid of spectral standard stars.

To obtain accurate spectral classifications for our program stars, we obtained a grid of optical spectra of F-type MK spectral standard stars with the same telescope and setup as our program stars. Our choices for standards are listed in Table 2. Although there are many choices for spectral standards, ours were chosen based on a careful consideration of the consistency of previous classifications and accessibility from the SMARTS 1.5 m telescope at CTIO.

Unfortunately, A8V and F4V spectral standards do not appear in the literature (e.g., Morgan & Keenan 1973; Gray & Corbally 2009), and the only A9V standard that we could find in the literature (44 Cet = HR 401; Gray & Garrison 1989) had several discrepant published spectral types. For subtypes A8V and A9V we adopted a star that had been assigned that spectral class by at least one expert classifier and had colors representative of other stars of the same classification. We adopted the Hyades member HD 27561 as an F4V standard. The star was originally classified as F4V by Morgan & Hiltner (1965), and its Hipparcos B − V color (0.41) is intermediate in color between the F3V and F5V standards classified by Morgan (and retained as standards by Morgan & Abt 1973, Morgan et al. 1978, and Gray 1989)—HD 26015 (F3V standard, B − V = 0.40) and HD 27524 (F5V standard, B − V = 0.434)—both of which are also Hyads. We have confirmed that its spectrum is intermediate between the F3V and F5V standards. For the A8V standard, we adopt the star HD 158352 (HR 6507), which was classified as A8V by Cowley et al. (1969), Abt & Morrell (1995), and as A8Vp by Mora et al. (2001). For the A9V standard, we adopted the Praesepe member HD 73450 (BS Cnc), based on its A9V classifications by Bidelman (1956), Abt (1986), and Gray & Corbally (2002). We verified that the adopted A8V and A9V "standards" had optical spectra that were morphologically intermediate between the A7V and F0V MK standards.

With our visual classification tool and a complete grid of dwarfs, we iteratively classified the 138 program stars at least four times before settling on a final spectral type. We conservatively estimate our classifications to be correct to within one subtype.

Spectral types for many of our candidate members are available through the Michigan Spectral Survey (Houk & Smith-Moore 1988; Houk 1982, 1978; Houk & Cowley 1975) which allows us to directly compare our newly obtained spectral types with past results from plate surveys. As Figure 1 shows, the newly obtained spectral types agree quite well with those obtained in the Michigan Spectral Survey. The average difference is small: 0.5 ± 1.1 (1σ). The Houk classifications are based on comparison to the "MK" standards from Johnson & Morgan (1953), whereas we have relied heavily on the "revised MK" F-type standards from Morgan et al. (1978, "MK78") (these later types have been adopted by Gray, Garrison, Corbally, and collaborators in their surveys). Any systematic differences in the classifications likely come from differences in the choice of standard star, which differs between us and Houk mostly among the earliest and latest F stars.

Zoom In Zoom Out Reset image size

To calculate individual extinctions in the Johnson–Cousins and 2MASS bands, we constructed a modern spectral type-intrinsic color calibration with B − V, V − I_c, V − K_S, J − H, and H − K_S colors using samples of nearby dwarf stars from the Hipparcos catalog (Perryman & ESA 1997), normal cool dwarfs from Neill Reid's compiled photometric catalog for stars in the 3rd Catalog of Nearby Stars,⁶ and a cross-matched catalog combining the new Gliese-2MASS catalog (Stauffer et al. 2010) and the catalog of average Johnson UBV photometry from Mermilliod (2006). For the Hipparcos sample, only dwarfs with luminosity class "V" and M_V within 1 mag of the Wright (2005) main sequence were selected, and only stars with parallaxes greater than 13.33 mas and relative parallax error less than 12.5% were included (d < 75 pc), to select those stars ostensibly within the "Local Bubble" which has negligible reddening. An intrinsic color sequence was constructed for B- through M-type stars, and the adopted intrinsic colors for a given dwarf spectral subtype were adopted based on a best B − V color.⁷ Our adopted intrinsic colors are in Table 3.

Many of the stars in our sample had very low extinction, and where we obtained a non-physical negative extinction we set the extinction to zero. We used a total-to-selective extinction of R_V = 3.1 and calculated A_V using the color excesses E(B − V), E(V − I_c), E(V − J), E(V − H), and E(V − K_S) with the extinction ratios $A_{I_c}$/A_V = 0.58, A_J/A_V = 0.27, A_H/A_V = 0.17, $A_{K_S}$/A_V = 0.11 (Fiorucci & Munari 2003) to estimate extinctions for each star. We adopted the median A_V with the standard deviation as a conservative estimate of the uncertainty. These are listed in Table 4 with our other derived stellar parameters.

While the stars in our sample have been selected from the Hipparcos catalog and therefore have trigonometric parallaxes, we can refine the distance determination by employing moving cluster or "convergent point" parallaxes. This works by leveraging the longer-baseline astrometric measurements employed to determine proper motions and taking advantage of the common space motion for the group. Kinematically improved parallaxes have been shown to reduce scatter in the H-R diagram (see the Hyades in de Bruijne et al. 2001 for a good example) and a comparison with trigonometric parallaxes can further help identify interlopers. One caveat is that kinematic parallaxes are meaningless for stars which are not true members.

For our kinematic parallax calculations we followed Mamajek (2005) and used updated space motions for US, UCL, and LCC tabulated in Chen et al. (2011).⁸ As described previously, we use proper motion data from the revised Hipparcos reduction (van Leeuwen 2007) for stars with single star (five-parameter) solutions and data from Tycho-2 (Høg et al. 2000) otherwise. Our kinematic parallaxes compare very well with the Hipparcos trigonometric parallax data (Figure 2).

Zoom In Zoom Out Reset image size

While the techniques used to define the sample in de Zeeuw et al. (1999) ensure that the candidate members are physically coincident with Sco-Cen and moving with similar space motion, the selection is purely based on astrometric information. Here we attempt to use our spectroscopic observations in conjunction with an additional kinematic criterion to identify interlopers.

While the strength of Li can be used as a youth indicator in G- and K-type stars, older F-type stars will show less Li depletion and thus we cannot rely on this spectral feature alone as a youth indicator in our sample (Balachandran 1991, however see Chen et al. 2011). We can, however, use Li to determine membership for any G/K stars which are companions to F-type stars, excluding or confirming the pair as members. Since our sample should consist of stars with predominantly dwarf and subgiant like gravities, we can also identify and exclude any giants found in our sample.

For members, the kinematic parallaxes should be in agreement with the trigonometric parallaxes. As a conservative criterion, we reject any star as a member if the difference between trigonometric parallax and kinematic parallax exceeds three times the uncertainty in those quantities added in quadrature. The motivation is that if the disagreement between the trigonometric and kinematic parallaxes is significant at the 3σ level, the object clearly has a different space motion than the group and can be safely rejected.

4.4.1. US Interlopers

4.4.2. UCL Interlopers

4.4.3. LCC Interlopers

We can examine our red optical spectra for evidence of accretion with the Hα feature. Accreting stars are thought to produce Hα emission due to hot infalling gas. For our accretion criterion we use the Hα full width at 10% peak (W₁₀(Hα)). White & Basri (2003) find that independent of spectral type, a full width at 10% peak >270 km s⁻¹ is a good indicator of accretion. We do not estimate the mass accretion rates as that is beyond the scope of this study.

Using this criterion, we find only two stars in our sample with Hα emission consistent with accretion, both of which are binaries: the near-equal mass system AK Sco (HIP 82747; Andersen et al. 1989; Alencar et al. 2003) and the recently studied HD 101088 (HIP 56673; Bitner et al. 2010) with 10% peak widths of 710 km s⁻¹ and 400 km s⁻¹, respectively. The spectral resolution of our red optical spectra at Hα is ∼140 km s⁻¹. The Hα profiles of these accretors are shown in Figure 3 along with the Hα profile of a non-accreting star in our sample (HIP 67230, F5V) with the same spectral type as HD 101088 and AK Sco.

Zoom In Zoom Out Reset image size

Using the classical T Tauri star color–color locus as determined by Meyer et al. (1997) and the intrinsic color locus of main-sequence stars we look for near-IR photometric signatures of accretion among our Sco-Cen members using 2MASS near-IR photometry. The plot in Figure 4 shows our Sco-Cen members (without reddening corrections applied) clustered around the locus with spread consistent with the uncertainties in the colors. The one obvious outlier with H − K_S > 0.4 is the known near-equal mass binary AK Sco referred to above. AK Sco is the star to the right of the locus, and its position is similar to that of the Herbig Ae/Be stars in Hernández et al. (2005).

Zoom In Zoom Out Reset image size

Here we estimate the optical spectroscopic accretion disk fraction for F stars in our sample as 0/17 (<19%; 95% CL) for US, while UCL has 1/41 (2⁺⁵₋₁%; 68% CL) accretors and LCC has 1/50 (2⁺⁴₋₁%; 68% CL) F-type accretors. This compares well with the Carpenter et al. (2006) Spitzer results in which 0/30 (<11%; 95% CL) F- and G-type stars were found to be accretors. The F-type results in UCL and LCC are consistent with the lower-mass G- and K-type stars studied by Mamajek et al. (2002), in which 1/110 (0.9^+2.0_−0.3%; 68% CL) members exhibited spectroscopic evidence of accretion.

As mentioned previously, PMS stars may exhibit enhanced chromospheric activity, which will manifest itself through partially filled Hα absorption lines. To examine this effect we measured the strength of the Hα lines with IRAF⁹ for all our program stars as well as our spectral standards and plot the EW(Hα) against T_eff (Figure 5). The members plotted here indicate slightly enhanced chromospheric activity, although most are still within the 2σ spread of the spectral standards.

Zoom In Zoom Out Reset image size

In order to compare Sco-Cen members with theoretical models and explore the star formation history of the F-type members, we construct a theoretical H-R diagram. The adopted effective temperature (T_eff) scale and bolometric correction (BC) scale were taken from an extensive set of notes for spectral subtypes by EM.⁶ This new T_eff scale comes from careful review and inter-comparison of spectroscopic and photometric temperature estimates for high quality MK standards as well as large samples of field dwarfs with MK classifications. The bolometric corrections are consensus estimates for the adopted temperatures, which rely heavily on the BCs for hot dwarf stars from Bessell et al. (1998) and for cool dwarf stars on the series of papers by Casagrande et al. (2006, 2008, 2010) (however BCs were also calculated interpolating the scales of Code et al. 1976, Balona 1994, Flower 1996, and Bertone et al. 2004 for comparison). Our temperatures and bolometric corrections are listed in Table 3.

We combine our extinction estimates with our kinematic parallaxes and V-band photometry (Perryman & ESA 1997) to estimate bolometric luminosity and place the members on a theoretical H-R diagram. We compare our data to the Dartmouth PMS models (Dotter et al. 2008) in Figure 6.

Zoom In Zoom Out Reset image size

For stars in UCL and LCC earlier than ∼F5, the H-R diagram positions of the stars are very near the main sequence and thus we cannot reliably determine ages from PMS evolutionary tracks. Therefore, we only use F5 and cooler stars to estimate the median age of UCL and LCC. For all stars in US and those later than F5 in UCL and LCC we calculate ages and masses by linearly interpolating between isochrones from Dotter et al. (2008), Demarque et al. (2004), Siess et al. (2000), and D'Antona & Mazzitelli (1997). Figure 8 shows the distribution of ages for the three subgroups. Individual results for our ages and masses are listed in Table 4. For stars which are too close to the main sequence to infer a reliable age, we determine a lower limit on the age using the 2σ upper limit on the luminosity and estimate a mass by assuming an age of 15 Myr (10 Myr for US). Our median age estimates for each subgroup are listed in Table 6.

Zoom In Zoom Out Reset image size

Interestingly, we obtain a significantly older median age for Upper Sco than other studies of the low-mass members (5 Myr, Preibisch et al. 2002; 6–8 Myr, de Zeeuw & Brand 1985; 10 Myr, Glaspey 1971). As seen in Figure 8, we obtain a median age of 13 Myr for Upper Sco using the evolutionary tracks of Dotter et al. (2008) and an overall median age from all tracks of 13 ± 1 Myr (68% CL). This is disconcerting, especially considering that the median F-star ages we obtain for LCC and UCL are consistent with the main-sequence turnoff ages and the PMS G- and K-type stars examined in Mamajek et al. (2002). Motivated to resolve this large discrepancy, we revisit the Upper Sco main-sequence turnoff ages, the age of the M supergiant Antares, examine the ages of the A-type stars and the ages for the PMS G-type stars. We also examine a kinematic expansion age for Upper Sco.

5.1.1. Intrinsic Age Spreads

In order to constrain how much of the observed spread in ages is due to observational uncertainty and estimate the intrinsic age spread, we performed a simple Monte Carlo simulation. First we calculated the observed 1σ dispersion in ages for each subgroup by simply taking the standard deviation of the ages (F5 and later stars for UCL and LCC), with deviant outliers clipped using Chauvenet's criterion (Bevington & Robinson 2003). These observed spreads are shown in Table 6.

For each subgroup, we generated a synthetic population of 10⁴ H-R diagram positions using the median H-R diagram position and median uncertainties for our member stars (F5 or later only for UCL and LCC), assuming a Gaussian distribution of errors. We then obtained ages and calculated the 1σ spread in ages for these three synthetic populations, again clipping deviant outliers with Chauvenet's criterion since the same criterion was applied to our real data to obtain the 1σ dispersion in ages. The distribution of ages for these simulated populations with the Dotter et al. (2008) evolutionary tracks is shown in Figure 9.

Zoom In Zoom Out Reset image size

Our estimates here assume that the observed dispersion in ages is the result of the true dispersion in ages and the dispersion in ages as a result of the observational uncertainties, i.e., σ²_observed = σ_intrinsic² + σ²_{uncertainties}. Using the Dotter et al. (2008) models, we then estimate the intrinsic dispersion in ages using the dispersions from our real and simulated populations. Rather than give preference to one set of evolutionary models, we repeat this calculation for each of our evolutionary models. This way we obtain a range of intrinsic age dispersions.

For US the observed 1σ dispersion in ages ranges from ±3 Myr with the D'Antona & Mazzitelli (1997) models to ±5 Myr with the Siess et al. (2000) models. When we account for the age dispersion from the observational uncertainties, we find the intrinsic 1σ age dispersion ranges from ±1 Myr to ±3 Myr, again depending on the models. For UCL and LCC, which have an observed age dispersion of ±5–9 Myr and ±5–9 Myr (depending on models), we find intrinsic age dispersions of ±4–7 Myr and ±0–8 Myr, respectively. This indicates that 68% of the star formation in Sco-Cen occurred over a time span of 2–6 Myr (US), 8–14 Myr (UCL), and <16 Myr (LCC). These are upper limits as we have not accounted for stellar binarity. Our findings show a much smaller age dispersion for US than the other two subgroups, consistent with the smaller age spreads found by Preibisch et al. (2002) and Slesnick et al. (2008), who also found a very small intrinsic dispersion in age. For UCL and LCC our age spreads are larger than that found in Mamajek et al. (2002), who found that 68% of the star formation had occurred within 4–6 Myr for UCL and LCC.

The most recent age determination for the turnoff was performed by de Geus et al. (1989) using Walraven photometry in which they estimated a turnoff age of 5 Myr using Maeder (1981) evolutionary tracks with overshooting but no mass loss or rotation. Preibisch et al. (2002) examined the H-R diagram again and argued that the data were consistent with an age of 5 Myr with very little spread in ages. We now have updated stellar evolutionary models and Hipparcos astrometry which we employ to revisit the turnoff age. We use the early-type members from de Zeeuw et al. (1999) with the addition of δ Sco, a long-period binary that is certainly a member of US but was not a de Zeeuw et al. (1999) member of US due to its perturbed motion. To construct the H-R diagram, we take Strömgren photometry from Hauck & Mermilliod (1998) and de-redden it according to the prescription of Shobbrook (1983). We then use the T_eff and BC_V calibration of Balona (1994) with the de-reddened Strömgren photometry. Since we are using the spread in isochrones near where the stars evolve off the main sequence, we restrict ourselves to stars hotter than log(T_eff) > 4.40 dex or more luminous than log(L/L_☉) > 4.00 dex. For stars cooler or less luminous than this the derived ages are very sensitive to the observational uncertainties and we are unable to estimate meaningful ages. Our derived temperatures and luminosities are given in Table 7. A plot of the data, with and without corrections for binarity (see below), is shown in Figure 10 with the evolutionary tracks of Bertelli et al. (1994).

Zoom In Zoom Out Reset image size

We have not used the runaway star ζ Oph in our analysis, for several reasons. First, while Hoogerwerf et al. (2000) used Hipparcos astrometry to conclude that ζ Oph most likely originated from Upper Sco ∼1 Myr ago, they note that it could have also originated from UCL ∼3 Myr ago. Unfortunately, the radial velocity of this star is poorly constrained¹⁰ and thus the UVW space motion is not sufficiently constrained to associate it with Upper Sco with great confidence. More importantly, however, a close examination of the abundance patterns in ζ Oph indicates that it has an anomalously high helium abundance (Herrero et al. 1992) and thus likely participated in a mass transfer event, perhaps from a close former binary companion. For this reason it is not reliable to use it to determine the turnoff age since it may have received extra nuclear fuel unaccounted for by the evolutionary models.

With only six B-type stars determining the turnoff age, it is important to assess the binarity of these stars since it may significantly alter the H-R diagram positions and, therefore, the derived ages. Only two stars, ω Sco and τ Sco, do not have a detected companion. β¹ Sco is a binary with ΔV = 1.26 ± 0.17 (Elliot et al. 1976). This magnitude shift, when accounted for, moves the primary down on the H-R diagram ∼0.1 dex in luminosity and gives an age for the primary of 9 Myr. π Sco is a binary with ΔV  ∼ 3.7 (Stickland et al. 1996), which moves the primary down on the H-R diagram ∼0.01 dex in luminosity, having virtually no effect on the derived age for π Sco. δ Sco is a binary with ΔV = 1.5–1.9 (Tango et al. 2009), shifting the primary down ∼0.07–0.1 dex in luminosity. However, because of its position it effectively moves parallel to the 9 Myr isochrone and thus does not affect the derived age for the primary. σ Sco is a quadruple system discussed extensively by North et al. (2007). They report the system as a spectroscopic pair (which includes the primary), a tertiary B7 component, and a B9.5V common proper motion companion separated by 20''. They deconstructed the spectroscopic pair into a B1III primary with a B1V secondary with ΔV ≃ 0.80, and a system age of ∼10 Myr using the Claret (2004) evolutionary models. Correcting our derived H-R diagram position for the magnitude of the secondary, we find that the primary moves down ∼0.2 dex, which changes our derived age to 8 Myr for the primary.

Unfortunately, the massive stars in US do not trace a well-defined turnoff, even with corrections for known companions. Including the shifted H-R diagram positions when accounting for the binarity, we estimate the turnoff age from the following stars: ω Sco (2 Myr), τ Sco (2 Myr), β¹ Sco (9 Myr), δ Sco (9 Myr), π Sco (12 Myr), and σ Sco (8 Myr). The median age for these stars is 9 ± 3 Myr (68% CL) with 1σ dispersion 4 Myr, midway between the previous age estimates of ∼5 Myr and our value of ∼13 Myr from the F-type members.

While most current stellar evolutionary models have ignored rotation, theoretical studies have shown that a moderate rotational velocity of v_eq  ∼ 200 km s⁻¹ will increase main-sequence lifetimes about 20%–30% (Meynet & Maeder 2000; Maeder & Meynet 2000; Talon et al. 1997). This represents a significant difference in ages obtained with evolutionary tracks ignoring rotation versus those which include a treatment of rotation. For this reason, we also considered the evolutionary tracks of Ekström et al. (2011) which include tracks for stars with and without rotation. It should be noted that neither the Bertelli et al. (1994) models nor the Ekström et al. (2011) models considered here include PMS evolution, but since the PMS evolution for a ∼9 M_☉ star is ∼10⁵ yr (Iben 1965) the PMS time can be safely neglected. Plotted in Figure 11 are the H-R diagram positions of the US main-sequence turnoff stars along with isochrones generated from the evolutionary tracks of Ekström et al. (2011) with no rotation (dashed lines) and those with a rotational velocity equal to 40% of the breakup velocity. The median age with rotation at 40% of breakup is 10 ± 2 Myr (68% CL) while the median age without is 9 ± 2 Myr (68% CL). Since the US turnoff stars are rotating with a median projected rotational velocity of 〈v sin i〉 = 96 km s⁻¹ (see Table 7) and 〈v_eq〉 = (4/π)〈v sin i〉, then the turnoff stars have 〈v_eq〉 ≃120 km s⁻¹. Using mass estimates from the Bertelli et al. (1994) tracks, we estimate the median break-up velocity of the US turnoff stars as ∼450 km s⁻¹ and thus the median observed rotational velocity is ∼25% of breakup. This is a significant median rotational velocity and indicates that rotation should be considered. Therefore we adopt the median age obtained with the rotating evolutionary tracks of 10 ± 2 Myr (68% CL).

Zoom In Zoom Out Reset image size

Our Upper Sco turnoff age is much older than the age derived by the de Geus et al. (1989) study. We believe that the major source of discrepancy between the de Geus et al. (1989) US turnoff ages and our US turnoff ages is the updated evolutionary tracks. If we use the de Geus et al. (1989) H-R diagram positions with the Bertelli et al. (1994) evolutionary tracks, we obtain a median age of 9 Myr, the same age we obtain with our modern data (though our individual H-R diagram positions are not the same as those in the de Geus et al. 1989 study). We also note that the use of evolutionary tracks which account for rotation tends to increase the derived ages by ∼25% in general (Meynet & Maeder 2000). For example, using the Ekström et al. (2011) rotating evolutionary tracks for a typical upper main-sequence star in LCC and UCL gives an age ∼30% older than ages derived with the Bertelli et al. (1994) evolutionary tracks. Revisiting the turnoff ages for UCL and LCC is beyond the scope of this paper, but may be considered in a future paper.

Antares (α Sco) is a rare M1.5Iab-Ib supergiant (Keenan & McNeil 1989) in US. As the most massive secure member of US, it places tight constraints on the age of the group. To estimate the H-R diagram of Antares we use the updated temperature scale of Levesque et al. (2005) for red supergiants, together with the revised Hipparcos parallax (van Leeuwen 2007) of π = 5.89 ± 1.00 mas and the angular diameter measurement of 41.3 ± 1.0 mas (Richichi & Lisi 1990). We obtain log(T_eff) = 3.569 ± 0.009 dex (assuming spectral type uncertainty of 1 subtype), log(L/L_☉) = 4.99 ± 0.15 dex. This H-R diagram position is plotted in Figure 12. With this H-R diagram position we obtain an initial mass of 16.6 M_☉ and an age of 11⁺³₋₁ Myr with the evolutionary tracks of Bertelli et al. (1994). However, just as with the main-sequence turnoff stars, the main-sequence lifetime of Antares was likely significantly altered by rotation and therefore we again consider the rotating evolutionary tracks of Ekström et al. (2011), shown in Figure 13. From these rotating evolutionary tracks we obtain an initial mass of 17.2 M_☉ and an age of 12⁺³₋₁ Myr, which we adopt as our final age for Antares.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

However, independent of this H-R diagram position we can still place constraints on the age of Antares using only the T_eff derived from the spectral type. Using the previously mentioned T_eff calibration for supergiants, a T_eff of 3710 K (log(T_eff) = 3.569) means that the age must be 9 Myr or more, since younger isochrones are not predicted to reach temperatures that low. This is illustrated in Figure 12 with the placement of the 5 Myr isochrone to the left.

We place the A-type stars of US on the H-R diagram as another independent age indicator. The F-type stars do not appear to be reaching the main sequence and thus we expect that some A-type stars will be PMS and therefore useful as an age indicator.

We use the kinematically selected A-type US members of de Zeeuw et al. (1999) and perform essentially the same analysis as used on the F-type stars. We estimate individual reddenings using the procedure described in Section 4.2. We calculate a kinematic parallax, again using the revised Hipparcos astrometry if the solution is that of a single star and use the Tycho-2 proper motions otherwise. We reject one star, HIP 77457, since its trigonometric parallax of 8.58 ± 0.86 mas disagrees with the kinematic parallax of 5.10 ± 0.42 mas by more than 3σ. Our individual H-R diagram position data for the US A-type stars are listed in Table 8 and plotted in Figure 14. As with the F-type stars, we do not see the A-type stars clustered around the 5 Myr isochrone, but rather mostly between the 5 Myr and 12 Myr isochrones.

Zoom In Zoom Out Reset image size

While the A-type stars in US have individual H-R diagram positions that are too close to the main sequence to yield reliable ages, we can examine the trend of the empirical isochrone and determine what ages are consistent with the ensemble. Shown in Figure 14 is the empirical isochrone, which we use to estimate the main-sequence turn-on for Upper Sco at spectral type ∼A3 or log(T_eff) ≃ 3.93 dex. We compare our empirical isochrones with the Dotter et al. (2008) evolutionary models, noting that an age of 8 Myr is too young to be consistent with the clump of stars at spectral type A3, and an age of 12 Myr is too old to be consistent with the clump of stars at spectral type A8–A9. Using these constraints, we estimate an age of 10 ± 2 Myr with the Dotter et al. (2008) evolutionary tracks. We obtain similar age estimates with other evolutionary models, summarized in Table 9. For the A-type stars we adopt the median age among all four evolutionary models considered of 10 ± 1 ± 1 Myr (statistical, systematic).

Since our F-type median ages for LCC and UCL agree very well with previous results obtained with the G-type PMS (Mamajek et al. 2002), we wish to directly compare our new results for US with a similar sample of G-type stars in an attempt to resolve this discrepancy in ages. We use the X-ray-selected G-type stars from Preibisch & Zinnecker (1999) and the kinematically selected G-type stars from de Zeeuw et al. (1999). Ages for the X-ray-selected objects from the Preibisch & Zinnecker (1999) study had been estimated previously using UKST Schmidt plate photometry. However, more precise Tycho-2 and 2MASS photometry are now available and thus we may be able to obtain a more reliable age estimate. Ages for the US G-type stars in the de Zeeuw et al. (1999) sample have not been previously examined.

For the X-ray-selected sample we cross-reference the Preibisch & Zinnecker (1999) sample with the Tycho-2 catalog (Høg et al. 2000) and select objects which have proper motions within the de Zeeuw et al. (1999) proper motion boundaries for Upper Sco (i.e., 0 mas yr⁻¹<μ < 37 mas yr⁻¹). This leaves us with 10 X-ray-selected G-type US members. We correct for extinction using (B − V), (V − J), (V − H), and (V − K) colors from Tycho-2 and 2MASS photometry, with the Tycho-2 photometry converted to Johnson using the conversions found in Mamajek et al. (2002, 2006). To estimate bolometric luminosity we use the converted V-band photometry, kinematic distances using the Tycho-2 proper motions and temperatures and bolometric corrections listed in Table 3. For the kinematically selected sample, we first identify two interlopers, HIP 83542 and HIP 81392. Houk & Smith-Moore (1988) classified HIP 83542 as G8/K0 III, a giant, and the H-R diagram position is well above the 1 Myr isochrones, consistent with this assessment. HIP 81392 and HIP 83542 both exhibit weak lithium absorption features, significantly lower than the other six kinematically selected US candidates from de Zeeuw et al. (1999) and inconsistent with the youth of Upper Sco (E. Mamajek 2011, private communication). With these interlopers removed, we are then left with six G-type US members from de Zeeuw et al. (1999). For the de Zeeuw et al. (1999) G-type US candidate members we estimate extinction, T_eff, distance, and bolometric luminosity using the same method as for the F-type stars as previously discussed. Our newly derived extinctions, distances, and H-R data are listed in Table 10, with the H-R diagram for this analysis shown in Figure 15.

Zoom In Zoom Out Reset image size

For the 16 G-type members of Upper Sco described above, we find median ages of 8–14 Myr using the evolutionary tracks of Dotter et al. (2008), Demarque et al. (2004), Siess et al. (2000), and D'Antona & Mazzitelli (1997). Our derived ages for these G-type stars of Upper Sco are summarized in Table 11, with an overall median age among all tracks of 9 ± 2 ± 3 Myr (statistical, systematic). The kinematically selected members are slightly more biased toward younger ages than the X-ray-selected members. However, this slight difference may be attributed to the brightness limits of the Hipparcos catalog. The difference may not be significant, though, with such small numbers of members.

One of the earlier published ages for US is an expansion age of ∼5 Myr derived by Blaauw (1978). Deriving ages using kinematic data holds obvious appeal since it does not depend on stellar evolution models. However, Brown et al. (1997), performing simulations of expanding OB associations and analyzing the resulting simulated proper motion data, found major problems with kinematic expansion ages which only rely on proper motion data. They found that expansion ages determined by tracing the proper motion of the stars back to their smallest configuration, as was performed in Blaauw (1978) to obtain the often quoted age of 5 Myr for US, always lead to underestimated ages and that all age estimates converge to ∼4 Myr. Furthermore, they found that the initial size of the OB association provided by this method is always overestimated. The Brown et al. (1997) study found that kinematic expansion ages determined using proper motion data alone are essentially meaningless.

Radial velocities must be considered in order to derive a meaningful expansion age. Given the availability of good quality trigonometric parallax and radial velocity data and the findings of Brown et al. (1997), we revisit the expansion age for US here. Following Mamajek (2005), we adopt a Blaauw expansion model (Blaauw 1956, 1964) which gives the observed radial velocity for parallel motion as

$$\begin{equation*} v_{{\rm obs}} = v^{\prime }\cos {\lambda } + {\kappa }r + K, \end{equation*}$$

where v' is the centroid speed of the group, λ is the angular distance to the convergent point, v'cos λ = v_pred is the predicted radial velocity with no expansion, κ is the expansion term in units of km s⁻¹ pc⁻¹, r is the distance to the star in pc, and K is an offset term which may include intrinsic effects (e.g., convective blueshift or gravitational redshift). We then write the difference between the predicted and observed radial velocities as

$$\begin{equation*} v_{{\rm obs}} - v_{{\rm pred}} = {\Delta }V_R= {\kappa }r + K \end{equation*}$$

so that κ is the slope in the plot of ΔV_R versus r:

$$\begin{equation*} \kappa = \frac{d({\Delta }V_R)}{dr}. \end{equation*}$$

The expansion age in Myr is then

$$\begin{equation*} \tau = \gamma ^{-1}\kappa ^{-1}, \end{equation*}$$

where γ = 1.0227 pc s Myr⁻¹ km⁻¹, a conversion factor.

To determine the expansion age we have started with the entire de Zeeuw et al. (1999) sample for which radial velocity measurements were available in the literature (listed in Table 12), since these all have individual trigonometric parallaxes. We have used radial velocities from Dahm et al. (2011), Chen et al. (2011), Gontcharov (2006), and Duflot et al. (1995). We limited the sample to those de Zeeuw et al. (1999) candidate members which have revised Hipparcos distances (van Leeuwen 2007)¹¹ within 50 pc of the mean US distance (145 pc; de Zeeuw et al. 1999). We also excluded candidate members with radial velocities more than 20 km s⁻¹ discrepant from the predicted radial velocities, as these are almost certainly non-members or unresolved binaries. Our v_pred is calculated using the new best estimate of the mean UVW motion of Upper Sco, detailed in Chen et al. (2011). The one-dimensional velocity dispersion in US is 1.3 km s⁻¹ (de Bruijne 1999b). In order to determine if there is evidence of expansion, we plot the distance r versus ΔV_R in Figure 16. Expansion would be exhibited by a positive correlation between the distance and ΔV_R, characterized by the slope (κ) of a line fit to the data. To take into account the observational errors in distance and radial velocity in fitting a line to the data, we performed a Monte Carlo simulation in which we added random Gaussian errors commensurate with each observational error to the observed data point, and then we fit a line using an unweighted total least squares algorithm. We performed 50,000 trials and then used the median and 1σ spread to quantify the effect of the uncertainties in the observational data on the slope. This yielded a Pearson r of −0.03 ± 0.11 and a best-fit line with a slope of κ = −0.01 ± 0.04 km s⁻¹ pc⁻¹. If we consider only models of expansion (i.e., κ > 0), then the 99% confidence lower limit on the expansion age is 10.5 Myr. However, the results are statistically consistent with no expansion. For an expansion age of 5 ± 2 Myr we would expect κ = 0.20^+0.13_−0.06 km s⁻¹ pc⁻¹, and an expansion age of 10 ± 2 Myr would have κ = 0.10  ±  0.02 km s⁻¹ pc⁻¹. Given that we are only able to obtain a lower limit on the expansion age for US and our expansion data are statistically consistent with no expansion, we do not consider our 99% confidence lower limit in the final age determination for US.

Zoom In Zoom Out Reset image size

Every age indicator we have examined thus far has yielded an age of Upper Sco of 9–13 Myr, summarized in Table 13. The youngest median age comes from the G-type stars at ∼9 Myr, and the oldest from the F-type stars at ∼13 Myr, so we can conclude with confidence that US must be between 9 and 13 Myr old. Regarding each segment of the H-R diagram as independent, we simply take the mean age of our values—∼10 Myr for the main-sequence turnoff, ∼12 Myr for Antares, ∼10 Myr for the A-type stars, ∼13 Myr for the F-type members, and ∼9 Myr for the G-type members. This gives a final adopted age of 11 ± 1 Myr (s.e.m., statistical) for Upper Sco, which is more than twice the currently accepted age (Preibisch et al. 2002). As each segment of the H-R diagram provides an independent age, we estimate our systematic uncertainty as the 1σ dispersion of these independent ages, which yields a systematic uncertainty of ±2 Myr.

Upper Sco has been a benchmark stellar population in studies of circumstellar disk lifetimes (e.g., Carpenter et al. 2006), but has a lower primordial disk fraction when compared to other stellar populations at 5 Myr (e.g., λ Ori; Barrado y Navascués et al. 2007). Considered in the context of 11 Myr age, the observed primordial disk fraction is consistent with those of similarly aged stellar populations (e.g., NGC 7160; Sicilia-Aguilar et al. 2006). Our detailed analysis of the isochronal ages for massive and intermediate-mass Upper Sco stars may indicate that the ages of similarly aged groups may also be due for further investigation and significant revision. Naylor (2009) has also found evidence that the nuclear ages for massive stars in <10 Myr old groups are typically 1.5–2× the pre-MS contraction ages. In particular, Naylor (2009) finds that two of Upper Sco's ∼5 Myr old siblings (NGC 2362 and Cep OB3b, each with pre-MS ages of 4.5 Myr) have best-fit isochronal ages for the massive stars of 9.1 Myr and 10 Myr, respectively. In this regard the doubling of Upper Sco's age does not seem surprising. Naylor (2009) has proposed that the nuclear ages may be the more correct ones, and that protoplanetary disks may correspondingly have nearly twice as long to form gas giant planets as previously believed.

In addition to circumstellar disk studies, Upper Sco has been the subject of many surveys for low-mass companions (e.g., Lafrenière et al. 2008; Ireland et al. 2011). A low-mass companion to [PZ99] J160930.3-210459¹² was discovered at a very large separation by Lafrenière et al. (2008). The mass¹³ of this companion was determined to be 8 M_Jup using the assumed age of 5 Myr. However, we use our revised age for Upper Sco of 11 Myr and the object's reported T_eff of 1800 ± 200 K (Lafrenière et al. 2010) to obtain a mass of 13⁺²₋₃ M_Jup using the DUSTY models (Chabrier et al. 2000; Baraffe et al. 2002). Alternatively, using the object's bolometric luminosity of log(L/L_☉) = −3.55 ± 0.2 (Lafrenière et al. 2010) and the 11 Myr age we estimate a mass of 14⁺²₋₃ M_Jup, consistent with the estimates from T_eff. Here we follow Lafrenière et al. (2010) and adopt the estimates from the luminosity, listed in Table 14. Similarly, a companion to GSC 06214-00210 was discovered by Ireland et al. (2011) and thought to have a mass of ∼14 M_Jup at an age of 5 Myr. However, at an age of 11 Myr the mass is slightly higher at 17 ± 3 M_Jup, using the DUSTY models and the estimated absolute magnitudes M_J ∼ 10.5, M_H ∼ 9.6, M_K ∼ 9.1 (Ireland et al. 2011). The substellar companion to HIP 78530 (Lafrenière et al. 2011) was estimated to have a mass of ∼21–26 M_Jup with an assumed age of 5 ± 1 Myr. We use the bolometric luminosity of log(L/L_☉) = −2.55 ± 0.13 (Lafrenière et al. 2011) and our revised age of 11 Myr to obtain a mass for HIP 78530B of 30⁺¹⁷₋₈ M_Jup. The substellar companion to the young brown dwarf UScoCTIO 108 has a mass of 14⁺²₋₈ M_Jup with an assumed age of 5–6 Myr (Béjar et al. 2008). Using the reported luminosity of log(L/L_☉) = −3.14 ± 0.20 (Béjar et al. 2008) and our revised age of 11 Myr, we estimate a mass of 16 ± 2 M_Jup for UScoCTIO 108b. The substellar binary Oph J1622-2405 initially had component mass estimates of ∼14 M_Jup and ∼7 M_Jup based on an assumed age of 1 Myr (Jayawardhana & Ivanov 2006). However, follow-up spectroscopic studies (Allers et al. 2007; Close et al. 2007; Luhman et al. 2007) found higher masses. Here we estimate the masses using the effective temperatures from Luhman et al. (2007) with our revised age of 11 Myr and find masses of 53⁺⁹₋₇ M_Jup and 21 ± 3 M_Jup for Oph J1622-2405A and Oph J1622-2405B, respectively, with the DUSTY models. This is in agreement with the results from Luhman et al. (2007), who used H-R diagram positions with the DUSTY models for their mass estimates. The revised mass estimates discussed above are summarized in Table 14. Hence, we believe that none of the substellar companions directly imaged in Upper Sco have inferred masses below the deuterium-burning limit.

Sco-Cen has been used as an example of triggered star formation, with supernova in UCL/LCC triggering star formation in US and supernova in US triggering star formation in ρ Oph (de Geus 1992; Preibisch & Zinnecker 1999). The UCL shell has an expansion velocity of 10 ± 2 km s⁻¹ and radius 110 ± 10 pc, suggesting it originated in UCL ∼ 11 Myr ago and passed through US about 4 Myr ago (de Geus 1992). However, given the revised age estimates for US, it does not seem likely that the passage of this superbubble would have triggered star formation in US since it would have arrived too late. However, this does not mean that star formation in US was not triggered by UCL, it simply means that the timescales and ages are not consistent with the arrival of the UCL shell in US as the triggering event. Similarly, the shell around US has an expansion velocity of 10 ± 2 km s⁻¹ and a radius of 40 ± 4 pc, consistent with an age for the shell of ∼4 Myr. At 10 km s⁻¹ it would have traveled the ∼15 pc from US to ρ Oph in about 1.5 Myr. With the revised 11 Myr age for US and ρ Oph ages of ∼2–3 Myr (Erickson et al. 2011), there is sufficient time for star formation in ρ Oph to have been triggered by the US shell.

Another important implication of our result is to drive down the inferred progenitor mass for the runaway young neutron star RX J1856.5-3754. Tetzlaff et al. (2011) claim that RX J1856 formed in Upper Sco and was ejected 0.5 Myr ago. Assuming a progenitor age of 5 Myr, Tetzlaff et al. (2011) predicted that the progenitor would have had a mass of ∼37–45 M_☉ and main-sequence spectral type of O5-O7. Smartt (2009) reviewed the observational and theoretical constraints on the supernovae and their progenitors, and suggests that progenitors with masses of >30 M_☉ almost certainly form black holes, not neutron stars. However, if the progenitor of J1856 was a 10.5 Myr old star when it exploded 0.5 Myr ago, then the progenitor was most likely a ∼18–20 M_☉ O9-type star (using the Ekström et al. 2011 tracks), with the remnant being either a neutron star or black hole (likely depending on the rotation or whether it was in an interacting binary; Smartt 2009). Indeed, if the empirically constrained set of supernova outcomes outlined by Smartt (2009) are correct, and if one adopts the previous age of 5 Myr, then the deceased high-mass stars in Upper Sco should have all turned into black holes. This would leave the "young" neutron star RX J1856 with a well-constrained distance and proper motion but without a plausible birth site. We conclude that the RX J1856 runaway scenario is more plausible if Upper Sco is 11 Myr old, and the progenitor would have been a much easier to form ∼18–20 M_☉ star rather than a rarer ∼37–45 M_☉ star.