A NEAR-INFRARED SPECTROSCOPIC SURVEY OF CLASS I PROTOSTARS

Ca ii has a well-known near-IR triplet line at 8498 Å, 8542 Å, and 8662 Å. Ca ii emission is likely produced in the protostellar magnetospheric infall of gas (Azevedo et al. 2006) and thus is a useful accretion tracer. If the lines are optically thick, the line flux ratios should be 1:1:1 since these lines are very close in wavelength. Detailed modeling by Azevedo et al. (2006) shows that the line ratios are typically 1.1–1.2, but can be as large as 1.5. High-resolution spectra of ∼90 T Tauri stars, and found that the shortest wavelength line of the triplet tends to have the highest peak and also tends to be the narrowest of the lines (Hamann & Persson 1990). White & Hillenbrand (2004) presented high-resolution spectra of these lines for a large sample of T Tauri and Class I stars, showing that the Ca ii line profiles range from being very narrow to having an FWHM up to 250 km s⁻¹.

Although only 20 Class I YSOs (18% ± 4%) showed Ca ii emission, 20/33 (61% ± 14%) targets with detectable flux in the I band have Ca ii emission, making Ca ii a very common (although rarely observed) emission feature for Class I YSOs. Being associated with mass accretion, and specifically magnetospheric infall, Ca ii emission is strongly associated with H i Brackett (Br) γ. All targets that have Ca ii emission also show emission from Br γ, but 20/25 targets with a detected continuum at the I band and Br γ emission also have Ca ii emission. Thus, ∼20% of targets have Br γ emission but no emission from Ca ii. In terms of EW, Ca ii is the strongest emission line observed from Class I protostars.

We also found that the EW ratios of the Ca ii IR triplet can deviate significantly from 1:1:1 line ratios. The mean ratio for EW₈₄₉₈/EW₈₅₄₂ = 1.1, but ranges from 1.8 to 0.8. Similarly, the mean ratio for EW₈₄₉₈/EW₈₆₆₂ = 1.3, but ranges from 1.9 to 0.8. Gamiero et al. (2006) observed a similar range of line ratios in the EWs of the Ca ii lines for DI Cep, showing that these ratios are also variable. Azevedo et al. (2006) note that variation in the gas temperature and mass accretion rate of their models do not completely account for this variability.

The He i line at 1.0833 μm is seen in both absorption and emission from many young stars. The line profile is sensitive to the kinematics of the stellar wind (Dupree et al. 1992). Edwards et al. (2006) presented observations of 38 He i line profiles from T Tauri stars observed at a spectral resolution of R = 25,000, and demonstrated that young stars have a wide diversity of inner wind properties. In comparison, our data have much lower spectral resolution (R = 1200, δv ∼ 250 km s⁻¹). Details of the line profiles of He i will be discussed in a future paper.

The He i line was detected in 35 of our Class I YSOs. The He i line was detected in 32% (35/110) of our Class I YSOs and it was detected in 52% (35/67) of the targets where there was enough flux at the J band to detect the continuum. Among the He i lines detected, 69% (24/35) of the He i lines show subcontinuum absorption, and emission is seen in 77% (27/35) of them. Note that objects with He i absorption can also simultaneously have He i emission. We see deep He i absorption from the three FU Orionis-like stars where we detected the J-band continuum. Data from T Tauri stars show that the wind accelerating region is very close to the stellar photosphere, and is possibly the stellar corona (Edwards et al. 2006). However, the flux from FU Orionis-like stars is likely dominated by light from the disk (Hartmann & Kenyon 1985), yet we still see strong He i absorption, suggesting that in these cases the wind originates from the disk. The He i features for all of the FU Orionis-like stars in the sample show only strong blueshifted absorption (i.e., no emission or non-blueshifted absorption). Thus, we propose that this is a possible way to indicate if a candidate star is an FU Orionis-like star. However, some other stars also have an He i feature that shows only blueshifted absorption despite not sharing other spectroscopic features with FU Orionis-like stars, so this criterion cannot be used alone.

In this analysis, we consider the [Fe ii] line at 1.644 μm and the H₂ line at 2.122 μm. Both [Fe ii] and H₂ are well-known tracers of winds via shock-induced emission. Emission from these species is often seen far from the stellar source in HH flows. The excitation mechanism for H₂ within a few hundred AU of a young star remained poorly understood until recently. Possible excitation mechanisms include UV (either from the accretion flow onto the star or from the star itself), X-rays from the star's corona, or shocks from the wind. Beck et al. (2008) found that the properties of the H₂ emission (e.g., the emission morphologies and spatial extent) are most consistent with shocked excitation from the wind rather than excitation of gas by radiation from the central star. [Fe ii] emission is a very useful probe of winds, especially in regions where the extinction is high enough to preclude optical observations (Bally et al. 2007). [Fe ii] emission probes high excitation temperature (11,000–12,000 K) winds, and in particular fast (>30 km s⁻¹) dissociative J shocks (Reipurth et al. 2000).

The study by Beck et al. (2008) was based on observations of six Classical T Tauri stars. We seek to determine if H₂ emission from Class I protostars is also shock excited. The [Fe ii] and H₂ lines are also often observed in the spectra of Class I protostars, with 44/107 targets having [Fe ii] emission and 47/110 having H₂ emission. Among the 44 Class I objects that show [Fe ii] emission, 35 (80% ± 13%) also have H₂ emission. The probability that a random sampling of 44 spectra from this data set would yield 35 or more objects with H₂ emission is less than 10⁻⁶. Considering the strong correlation between the presence of [Fe ii] and H₂ emission, we conclude that H₂ emission from Class I YSOs is also likely to be excited by shocks in winds.

Figure 6 shows that there appears to be a weak trend of increasing [Fe ii] EW with increasing H₂ EW. All targets with an H₂ EW less (stronger emission) than −3 Å also show [Fe ii] in emission, provided there was enough flux to observe the H-band continuum. With two exceptions, all targets with [Fe ii] EW less than −2 Å have H₂ emission. With one exception, only targets with high veiling have an [Fe ii] EW less than −7 Å.

Zoom In Zoom Out Reset image size

Figure 7 shows that the Br γ and H₂ EWs appear to be related by a weak (the correlation coefficient is 0.03) inverse relationship. Targets with strong Br γ tend not to have strong H₂ emission and vice versa. Only targets with high veiling have strong emission from H₂ (EW <−7 Å) or Br γ (EW <−9 Å). It may be possible that higher mass accretion rates onto the star suppress the H₂ excitation mechanism. However, since H₂ emission is believed to be shock excited in the wind (Beck et al. 2008), which is accretion driven, this scenario seems unlikely. Another possibility is that the H₂ line flux is not dependent on the mass accretion rate, and that the higher veiling associated with higher Br γ EWs reduces the observed H₂ EW. However, as stated above, only targets with high veiling have strong emission from H₂, contrary to this scenario. Increasing veiling will tend to push the EW values toward (0,0) in Figure 7 (and in Figure 6 as well). Although many data points are near the origin of the figure, they are predominantly targets with low veiling. We also note that no early-type star where Br γ is seen in absorption shows H₂ emission. Spectroscopic monitoring should show if the variability of the Br γ and H₂ lines is correlated or independent of each other.

Zoom In Zoom Out Reset image size

We observe that the line EWs for H₂ and [Fe ii] are stronger in targets with high veiling. Objects with high veiling might be expected to have greater Br γ luminosities since both are well-known accretion tracers. Since the wind is powered by accretion, it would also be expected that targets with high mass accretion would have stronger winds. However, if both Br γ and veiling are accretion tracers, then why is there an inverse relationship between H₂ and Br γ EWs? These correlations show that the H₂ and [Fe ii] lines, which are from the shocks in the wind, are not proportional to the instantaneous mass accretion rate (traced by Br γ). Since atomic hydrogen is only ionized very close to the star whereas H₂ traces colder more distant gas, the source of these lines are not co-located. Also, we speculate in Section 6.4 that it may be possible that the mass accretion rate is quite high in the cases where there is strong H₂ and [Fe ii] emission, but that the Br γ emission mechanism has collapsed resulting in relatively lower Br γ emission with increased mass accretion rate. We stress that this result is applicable to the region immediately around the central star, and not for an extended outflow. The width of our 05 slit corresponds to a range of ∼70 AU at the distance of the Taurus star-forming region and ∼230 AU at the distance of the Orion star-forming region.

We have found that 23% ± 5% (25/110) of our targets show the CO band heads in emission. This is very close to the value for HH sources found by Reipurth & Aspin (1997) of 20%. Targets that show CO in emission tend to have high veiling such that no photospheric absorption lines are apparent. Of the targets that show CO in emission, 91% ± 20% (21/23) have high veiling whereas 9% ± 6% (2/23) have low veiling. Although two stars have CO emission and low veiling, these two stars are both early-type stars, one of which has veiling that is low enough that the H i lines can be seen at shorter wavelengths but high enough to almost completely veil the Brackett series of lines (r_k = 8.8 for IRAS 05513−1024). Targets with CO emission more often have high veiling than the sample as a whole, for which 55% ± 7% (61/110) show high veiling.

We have found that the CO and Br γ features are closely related (the correlation coefficient of their EWs is 0.44). Figure 8 plots the EW of CO versus Br γ and is divided into three parts. Region A includes all objects where CO is seen in emission. CO EWs that are consistent with absorption from the photosphere of a dwarf star are included in region B, from early-type stars with no CO absorption to M4, the latest type in our sample (see Section 3). The CO absorption in region C is in excess of what is expected to be observed from the photosphere of a dwarf star. This figure shows that targets with CO in emission also always show Br γ emission (an important tracer of magnetospheric accretion) and almost always have high veiling (red symbols) as well. Figure 8 shows a general trend of greater CO emission with greater Br γ emission, but with significant scatter. Br γ is often seen in emission when the CO absorption is consistent with a dwarf stellar photosphere (region B) and when the veiling is low (black symbols). There is a trend of increasing Br γ emission with decreasing CO band head absorption, possibly due to CO and/or veiling emission increasing as the mass accretion rate increases. Notably, Br γ never shows significant emission when the CO is in absorption beyond what is consistent with the star's spectral type. This situation is most commonly seen among FU Orionis-like stars (see Section 6.4), which also have high veiling.

Zoom In Zoom Out Reset image size

We interpret this relationship between CO and Br γ as follows. When CO and Br γ are in emission and the veiling is high, the accretion rate is quite high. The surface of the disk is hotter than the disk midplane, accounting for the CO emission (Najita et al. 1996) and the lack of any absorption features from the disk. At lower accretion rates, Br γ is still seen in emission. The veiling is low enough that photospheric absorption bands dominate the CO feature. The low veiling suggests that the warm (>1000 K) emitting surface area in the disk is relatively low. In contrast, FU Orionis-like stars are believed to be experiencing a burst of mass accretion. Strong CO absorption is observed from FU Orionis-like stars. Muzerolle et al. (1998) empirically derived the well-used relationship between Br γ emission and mass accretion rate with observations of T Tauri stars. Although FU Ori-like objects have very high accretion luminosities and thus very high mass accretion rates (up to 10⁻⁴ M_☉ yr⁻¹; Hartmann 2000), they lack many of the usual mass accretion signatures commonly observed from T Tauri and many other Class I YSOs (e.g., Br γ or Paschen β emission). Br γ is not detected in some of the FU Ori-like objects (04073+3800, 18270−0153W, 20568+5217, and 22051+5848) and only very marginally detected in the rest, as seen in region C of Figure 8. These observations show that the established relationship between observed Br γ emission versus mass accretion rate is not valid for these objects. As such, FU Ori-like objects may accrete mass via a different mechanism than the magnetospheric funnel flows that produce the well-known Br γ emission in Class I and T Tauri stars with lower mass accretion rates. It is possible that Br γ emission is proportional to the mass accretion rate at relatively low mass accretion rates, but the proportionality breaks down as mass accretion rate increases to the point where at very high mass accretion rates there is no observed Br γ emission. We postulate that the magnetosphere, which shapes the accretion flow for classical T Tauri stars, collapses when the accretion rate is as high as found among FU Orionis objects. For further discussion of FU Orionis-like objects, see Section 6.4.

Figures 4 and 5 show the relationship between CO emission/absorption and Na i + Ca i. The FU Orionis-like objects in our sample lie in region E, below and to the right of the luminosity Class III stars. A number of objects have sufficiently high mass accretion to push CO into emission, and these objects are located in region F of Figure 5. Absorption from Na i + Ca i is photospheric. However, Na i has also been observed in emission in many of these cases when veiling is high, resulting in occasionally negative values for the EW of (Na i + Ca i). Ca i was never seen in emission.

CO emission and absorption trace gas at a moderate temperature (a few 1000 K) and high density (n_H>10⁹ cm⁻³), however there are several potential excitation mechanisms. Figure 7 in Calvet et al. (1991) shows that young stars with a mass accretion rate below ∼10⁻⁷ M_☉ yr⁻¹ can have CO in emission due to stellar radiation heating of the disk surface or CO in absorption from the stellar photosphere. If this model is correct, then we should expect our sample to have many YSOs that are accreting mass (judging from their Br γ emission), have low veiling (suggesting a low mass accretion rate), with some YSOs showing CO in absorption and others showing CO in emission. Although we observed many examples of YSOs with low veiling and CO in absorption, CO emission is usually only observed when the veiling is high and thus presumably when the mass accretion rate is high. Also, Figure 7 in Calvet et al. (1991) predicts that early-type stars should have CO emission when the mass accretion is as high as 10^−5.5 M_☉ yr⁻¹. Two of the eight early-type stars in this study have CO in emission, whereas 6/8 have no CO emission or possible CO absorption (four of these also have atomic hydrogen emission showing that they have a disk and are actively accreting mass). Both of the early-type stars with CO emission also have high veiling, suggesting that the presence of CO emission is correlated with high veiling and not with the spectral type of the central star. Calvet et al. (1991) also predicted that high mass accretion rates push the CO into absorption, as observed with FU Orionis-like stars. Martin (1997) proposed that the CO emission could be from the magnetospheric funnel flow of mass onto the star. If this hypothesis is correct, then multi-epoch observations should show that the CO band head flux should vary in step with other mass accretion tracers related to the funnel flow, such as Br γ or Ca ii emission. More recently, Glassgold et al. (2004) concluded that CO emission arises in a thick layer near the surface of the disk where atomic hydrogen collisionally excites the CO. Velocity-resolved observations can also help to determine the location of the source of the CO emission.

We first used 2MASS J-, H-, and K-band photometry of our targets to estimate the extinction to our targets. We used the values of A_λ/A_K from Table 1 from Nishiyama et al. (2009). With these values, we reddened the 2MASS colors of our targets to the T Tauri locus derived by Meyer et al. (1997). 2MASS did not detect some of our targets in the J or H band, so we were not able to estimate the extinction to all sources based on 2MASS photometry.

Although this method of estimating the extinction is very simple, it has several caveats. 2MASS observations cannot resolve close binary stars, so this method is not applicable to close binaries of comparable brightness. Scattered light from circumstellar material will make the object appear bluer and affect the derived extinction to the source. If an object is hidden behind an edge-on disk, then all of the observed flux may be scattered light. As noted above, it is very common for Class I YSOs to be associated with an IR reflection nebula. Table 1 flags objects where the K-band flux is likely to be dominated by scattered light (no point source is observed) or where scattered light could affect the near-IR colors (the object is associated with a reflection nebula), as judged from examining K-band images of their reflection nebulae in Connelley et al. (2007).

We used a simple procedure to model the observed spectrum of each protostar to estimate the extinction, veiling temperature, and the magnitude of the veiling. This program adds an IR excess (i.e., veiling emission, which we simulate as a single-temperature blackbody) to the spectrum of a star from the SpeX spectral library (Cushing et al. 2005; Rayner et al. 2009), then applies extinction to that sum using the reddening law described above. Naturally, the result of this procedure depends on the input spectral type. When possible, we used the best-fit spectral type found by our EW matching program; otherwise, we used an M2V spectral type since that is the most common spectral type among the stars in this sample. Each model (star plus veiling behind extinction) is compared to the observed spectrum and the rms of the fit is calculated. The code runs through a user-defined grid of veiling temperatures, veiling amplitude (r_k), and visual extinctions (A_v) to minimize the rms difference between the observed spectrum and the model spectrum. We used veiling temperatures from 200 K to 2000 K, which is approximately the formation temperature of chondrules and refractory inclusions in meteorites (Boss & Graham 1993; Alexander et al. 2007). Contour plots of the rms residual of the fit versus veiling temperature and extinction show the confidence intervals for the fitted parameters. An example of the output of our modeling code is shown in Figure 9. Our 1σ uncertainties for veiling temperature and extinction are determined by the amount by which these parameters have to change from the best-fit values for the rms of the model to be twice the best-fit rms. In many cases, we removed strong emission lines (most often atomic and molecular hydrogen, and CO) from the observed spectrum to improve the rms fit.

Zoom In Zoom Out Reset image size

The ability of our modeling code to precisely derive the extinction to the source varies greatly from target to target, and also depends on the spectral type of the template spectrum. This is reflected in the uncertainties presented in Table 2. We find that there is a degeneracy between the veiling temperature and the amount of extinction if the spectrum is dominated by a continuum that smoothly increases with wavelength, which is quite common among Class I YSOs. The amount of extinction is often very well constrained at a given veiling temperature, but good fits can often be found for a wide range of veiling temperatures and corresponding extinctions. We also find that among stars with late-type spectra, the code can often fit the observed spectrum using different amounts of veiling for different template spectra, and thus this procedure is a poor discriminator of spectral type.

Figure 10 shows the histograms for extinction based on both near-IR broadband colors and continuum modeling. Figure 11 shows that the extinction estimates based on near-IR colors tend to be ∼30% lower than the estimates based on continuum modeling. This discrepancy may be caused by the use of different photometric systems on very red objects, or scattered light from an associated reflection nebula affecting the observed colors more than the spectra. Another difference is that the 2MASS colors deredden to the T Tauri locus, whereas the continuum modeling routine more accurately takes into account the contribution of the stellar photosphere and veiling emission without scattering. Although it would not account for this systematic offset, the large uncertainties in the extinction calculated from continuum modeling mean that these estimates are often consistent with the extinction estimates based on the near-IR photometry. The median value of the extinction distribution based on continuum modeling is 12.7 A_v magnitudes, whereas the median value of the extinction distribution based on 2MASS photometry is 9.8 A_v magnitudes.

Zoom In Zoom Out Reset image size

Our continuum modeling program also allowed us to derive a distribution for the veiling temperature, which peaks near 1500 K. This result is likely influenced by wavelength coverage of our data, since the peak of a 1500 K blackbody is at 1.9 μm, near the middle of our wavelength coverage.

We found it very difficult to precisely constrain the amount of veiling (r_k) for most of the stars in the sample. For objects whose spectrum is dominated by veiling, we can only say that the veiling is high, likely greater than r_k ∼ 5 since r_k = 8.8 is the highest veiling value that we were able to meaningfully constrain. In the cases where the spectrum shows clear photospheric lines, these lines are used to constrain the veiling. In these cases, uncertainty in the estimate of the spectral type of the star limits the uncertainty of the veiling. Therefore, for much of the analysis regarding veiling, we have simply divided this sample into two groups: targets with "high" veiling and targets with "low" veiling, as defined in Section 4.

If the intrinsic line ratio of a given pair of emission lines is known, the observed line flux ratio can be used to calculate the extinction to the source. Lines that have the same upper state are particularly useful since the line flux ratio only depends on the transition probabilities to the two lower states. In this case, there are no assumptions about the emitting gas being in local thermodynamic equilibrium (LTE) or concerns that the two lines may be being emitted by gas with different properties (temperature, density, distance from the source, etc.). In other cases, it is reasonable to assume LTE, and thus the intrinsic line ratio can be calculated.

[Fe ii] emission lines at 1.644 μm and 1.257 μm are often seen in the spectra of young mass accreting stars. The ratio of these line fluxes has previously been used to estimate the extinction to the Cas A supernova remnant (Eriksen et al. 2009) and to make an extinction map toward an active galactic nucleus (Storchi-Bergmann et al. 2009) since these two lines also share the same upper level. Reipurth et al. (2000) also discussed the use of these lines to estimate extinction. Since these lines are forbidden, they should be optically thin. As such, this pair of lines is a useful tool to estimate the extinction to a wind source. The intrinsic line ratio is not yet well constrained, as noted by Eriksen et al. (2009). Eriksen et al. (2009) used a value of [Fe ii]_1.257/[Fe ii]_1.644 = 1.49, based on observations of P Cygni by Smith & Hartigan (2006). Storchi-Bergmann et al. (2009) used a value of 1.36 based on calculations (Nussbaumer & Storey 1988) verified by observations (Bautista & Pradhan 1998). For the purpose of this paper, we adopt a value of [Fe ii]_1.257/[Fe ii]_1.644 = 1.36.

We observed both the 1.257 μm and the 1.644 μm [Fe ii] lines in 25 targets. Our extinction estimation used the same reddening law as with all of the other methods we used. Generally, our extinction estimates based on the [Fe ii] line ratios are consistent with the estimates based on continuum modeling, taking into account the mutual uncertainties. In the case of IRAS 04286+1801, the [Fe ii] lines were the only reliable way to estimate the extinction to this target among the methods we used. In this case, the near-IR flux is dominated by scattered light, making the object appear bluer and thus leading to an underestimated extinction. Furthermore, being an FU Orionis-like object, there is no apparent flux from the photosphere or atomic hydrogen emission (see Section 6.4). Since the [Fe ii] emission is from the wind, the extinction calculated from the [Fe ii] line ratio may be different from the true extinction to the protostar itself. However, considering that the slit was 05 wide, the [Fe ii] emission in our spectrum is very near the location of the protostar, and not from a distant part of the outflow. We note that the 1.644 μm was occasionally on the edge of a comparably strong Br 12 absorption line or emission line. Such cases are designated with an asterisk in Table 2. Higher spectral resolution would be useful in separating these features to get a better flux measurement of the [Fe ii] line.

Zoom In Zoom Out Reset image size

The Br γ (2.16612 μm) to Paschen β (1.28216 μm) line ratio can be used as an extinction tracer assuming Case B recombination. In this case, the intrinsic line ratio is 5.75 ± 0.15 (Storey & Hummer 1995). Although most of our targets show some Br γ emission, the Pa β emission line was detected in only about one-third of them, usually because of high J-band extinction and as such there being no detectable continuum at the J band. We did not extract orders for which there was no continuum flux, so in these cases we did not measure the Pa β EW. In those cases where both lines were measured, we used them to calculate the extinction to the target, and these values are listed in Table 2.

In the case of embedded protostars, the H₂ v = 1–0 Q(3) (2.42373 μm) to S(1) (2.12183 μm) lines have been used (Beck 2007) since these two lines share the same upper state. Many of our Class I YSOs have strong H₂ emission, and we attempted to use these lines to estimate the extinction to these sources. We found that the extinction derived from these H₂ lines often greatly differed from the extinction estimates based on broadband colors or continuum modeling and was occasionally negative. We determined that this line ratio is an unreliable extinction estimator since in 4/10 cases, this line ratio was observed to be variable. By coincidence, the v = 1–0 Q(3) line is very close to a narrow (δλ = 7.1 Å) but opaque telluric absorption line at 2.42412 μm (in vacuum). The Doppler shift due to Earth's orbital motion moves this telluric line such that the v = 1–0 Q(3) line can be either on the blue wing of the line (and thus the Q(3) line would be partially absorbed) or completely clear of it. This telluric feature as well as the H₂ lines are not resolved at the spectral resolution (R = 1200) of our data. Thus, a significant amount of the flux of the v = 1–0 Q(3) line may be lost to this telluric absorption feature and not recovered when the telluric correction is carried out during the data reduction.⁵ This problem may be ameliorated by using sufficiently high spectral resolution to resolve this telluric absorption feature to accurately compensate for it. In the cases where the v = 1–0 Q(3) to S(1) line ratio did not vary, the observed line ratio was used to estimate the extinction with the understanding that the extinction may be underestimated if a significant amount of the flux of the Q(3) line is lost to this telluric absorption feature.

Having noticed that many of the stars with FU Orionis-like spectra (see Section 6.4) were associated with reflection nebulae, we investigated the connection between reflection nebulosity and the near-IR spectrum. We used the K-band images presented in Connelley et al. (2007) to classify the nebulae into three groups: strong nebulosity (no K-band point source is seen), some nebulosity (a K-band point source is seen along with a reflection nebula), or no nebulosity. The reflection nebulae have an average size of ∼10,000 AU or ∼20'' on the sky. The spectra as a whole were divided into three parts: spectra with "high" veiling (no photospheric lines apparent), spectra with "low" veiling (the spectrum shows photospheric absorption lines), and spectra that are similar to FU Orionis-like stars with deep water and CO absorption and without photospheric absorption lines.

The results of our analysis are gathered in Table 5. We found that although there are only 10 targets with FU Ori-like spectra (9% of the sample), 40% of the YSOs with strong nebulosity have this type of spectrum. Furthermore, all objects in our sample with FU Ori-like spectra are associated with a reflection nebula. An exception is IRAS 06297+1021W, which we consider to be an FU Orionis-like object despite having emission lines uncharacteristic of such an object, and which is not associated with a reflection nebula. YSOs with strong nebulosity have a higher fraction of FU Ori-like spectra (6/15) than objects without nebulosity (0/54) with greater than a 99.9% confidence. We found that the fraction of objects with high veiling has no dependence on the presence or absence of a reflection nebula. However, objects with low veiling are less likely to be found among targets with strong nebulosity than objects without nebulosity with a 99.9% confidence. These binomial confidence intervals were calculated using the expressions presented by Brandeker et al. (2006) in their Appendix B2. Targets without nebulosity and those with some nebulosity have similar fractions of objects with high veiling, low veiling, and FU Ori-like spectra, so these two nebulosity groups are difficult to distinguish based on their spectra. In summary, targets in this sample with strong nebulosity are less likely to have low veiling and more likely to be an FU Orionis-like star than targets with some nebulosity or no nebulosity.

These relationships show that there is a correlation between the appearance of the YSOs on large scales (>1000 AU) and the properties of the star and inner disk at very small scales. Objects with strong nebulosity could be objects with edge-on disks that block the central star from view, in which case we would see no K-band point source and all of the light we see is scattered off of circumstellar material. This is indeed the case for many of the objects observed by White & Hillenbrand (2004), who observed objects apparently more embedded than T Tauri stars (many objects were only seen in scattered light) and observed photospheric features in all of them. If objects with strong nebulosity are merely seen edge-on, then we would expect to observe the same fraction of objects with low veiling or FU Orionis-like spectra in all of the nebulosity groups. However, this is not what we observe. In order to not see a K-band point source, these objects must therefore be more deeply embedded on average, and presumably less evolved, than objects with some or no nebulosity. Among the objects with no reflection nebula, there may be no objects with FU Orionis-like spectra because either objects with no nebulosity are too evolved or because the increase in the luminosity of the object in the FU Orionis phase would illuminate nearby circumstellar material, creating a reflection nebula that previously was not visible.

Why are there so few objects with low veiling among objects with strong nebulosity, and so many objects with low veiling that have some or no nebulosity? The circumstellar material responsible for veiling also scatters and absorbs light, and can also obscure the central star. Objects with low veiling are most common among targets without a reflection nebula and quite rare among targets with strong nebulosity. If objects with low veiling were simply less luminous and thus do not illuminate a reflection nebula as often, then we would not see an equal number of low veiling objects among targets with some nebulosity and without nebulosity. One possible explanation is that objects with some or no nebulosity are simply more evolved, with less circumstellar material in the envelope (accounting for the optically thinner cloud, allowing us to see the central star) and less material in the inner disk (accounting for the lack of veiling emission). This would then suggest that FU Orionis objects, which are most common among targets with strong nebulosity, would be relatively less evolved.

The spectral type distributions of pre-main-sequence stars has been shown to have a strong dependence on the star-forming region (Luhman et al. 2003). Hillenbrand (1997) compared the spectral type distributions for the Ophiuchus, Chamaeleon, Taurus/Auriga, Lupus, L1641, and Orion Nebula Cluster (ONC) star-forming regions. Both the ONC and the Ophiuchus star-forming regions have a number of early-type stars, extending to OB in Orion and B type in Ophiuchus. However, the spectral type distribution of the ONC peaks at M3 whereas the spectral type distribution of Ophiuchus peaks earlier than M-class. The spectral type distribution for the Taurus star-forming region (e.g., Briceño et al. 2002; Luhman et al. 2003, 2006) peaks near K7, with most stars having spectral types from K0 to M8. While these studies found no stars in the Taurus star-forming region with a spectral type earlier than G, Kenyon & Hartmann (1995) found three young stars in the Taurus/Auriga clouds with spectral types earlier than G.

To determine if our sample of Class I YSOs shows a spectral type dependence versus star-forming region similar to the pre-main-sequence studies described above, we divided our sample into two groups of roughly equal size. Group 1 consists of targets in the Taurus, Auriga, and Perseus star-forming regions, and has 21 objects. These clouds are nearby, have a low stellar density, and are not forming massive stars. Group 2 consists of targets in the Orion and Ophiuchus clouds and also has 21 objects. The distributions of the best-fit spectral types are shown in Figure 3. The spectral type distributions are overall very similar. The two-sample Kolmogorov–Smirnov (K-S) test shows that the best-fit spectral type distributions have less than a 90% chance of being drawn from different parent populations. However, Group 1 has no star with a best-fit spectral type earlier than F5. In contrast, 6/21 (29%) of the stars in Group 2 are A or F stars. If the four known T Tauri stars are excluded, it remains true that Group 1 still has no star with a spectral type earlier than F5. Group 2 would then have 5/20 (25%) of stars with an A or F spectral type. The regions in which we found A- and F-type stars (Orion and Ophiuchus) are associated with well-known populations of early-type stars, in this case the ONC and the Sco-Cen OB association. We stress that these results are based on those stars for which we could determine a spectral type, and are only applicable to the clouds as a whole assuming that protostars of each spectral type are equally likely to have low enough veiling to allow a spectral type to be estimated. We also note that the Serpens clouds have very few targets for which a spectral type could be estimated. Of the 11 targets in that region, the spectra of only 2 targets have photospheric absorption lines, and we could get a spectral type estimate for only 1 target.

An important question to address is the fraction of Class I YSOs that are actively accreting mass. Unfortunately, the fraction of mass accreting T Tauri stars has not been well characterized, in part because the result depends on the age of the T association. The fraction of T Tauri stars with IR excesses, and thus warm disks, drops from ∼50% to ∼3% in a survey of star clusters ranging in age from 2.5 to 30 Myr (Haisch et al. 2001a; Hernández et al. 2008). Haisch et al. (2001b) found that 65% of young stars in IC 348 have disks, whereas the younger NGC 2264 and Trapezium clusters have disk fractions of 86% and 80%. Luhman et al. (2010) found that the disk fraction of the Taurus star-forming region (N(Class II)/N(Class II and III)) is ∼75% based on 348 objects observed by Spitzer. Using this metric, the results presented by Evans et al. (2009) show that the disk fraction in Chamaeleon II is 78% ± 18%, in Lupus is 66% ± 9%, in Perseus is 89% ± 6%, in Serpens is 81% ± 7%, and in Ophiuchus is 81% ± 6%. The presence of a disk does not mean that all of these stars are accreting matter, but only that they could.

We used the presence of Br γ emission (Muzerolle et al. 1998) or high veiling (when the veiling is high enough that no photospheric absorption lines are observed, r_k ≳ 5, characteristic of a warm inner disk close to the star) to determine the fraction of our targets that are accreting matter. Ninety-three stars show Br γ emission, whereas eight more have high veiling but no Br γ emission (these are mostly FU Orionis-like objects). Thus, 92% ± 9% (101/110) of Class I YSOs are actively accreting matter at any given time. Excluding the four known T Tauri stars, the fraction of Class I YSOs accreting matter remains 92% ± 9% (98/106).

The histogram of Br γ EWs in Figure 12 has been divided between targets with high and low veiling. Targets with high veiling have a mean EW of −4.21 and a median EW of −2.93, whereas targets with low veiling have a mean EW of −1.20 and a median EW of −1.63. EWs are measured relative to the continuum flux, which is elevated by the veiling in the case of targets with high veiling. Thus, targets with high veiling on average have much higher Br γ line fluxes and presumably high mass accretion rates as well. For comparison, the Br γ EW histogram for the T Tauri stars with Br γ emission observed by Muzerolle et al. (1998) has been overlaid in Figure 12. The two-sample K-S test shows that there is less than a 90% chance that the T Tauri Br γ EW distribution is drawn from a different parent population than Br γ EW distribution of the whole sample of Class I YSOs. Thus, the Br γ EW distributions for accreting T Tauri stars and accreting Class I protostars are indistinguishable with these data. We note that Muzerolle et al. (1998) selected T Tauri stars with previously measured mass accretion rates, so the targets they selected are not representative of T Tauri stars as a whole considering the large number of non-accreting weak line T Tauri stars. Although the Br γ EW distributions are consistent, Class I YSOs generally have higher veiling than T Tauri stars (Doppmann et al. 2005), so therefore Class I YSOs likely have a higher average Br γ line flux and presumably higher mass accretion rates.

Zoom In Zoom Out Reset image size

There are nine targets in this sample without Br γ emission or high veiling, and thus do not appear to be currently accreting matter. We have estimated the spectral types of eight of these. Among the non-accreting Class Is, early-type stars are over represented and late-type stars are under represented. The two-sample K-S test shows that there is a 90% chance that the spectral type distributions for accreting and non-accreting Class I YSOs are drawn from different parent populations. Three of these nine stars are A- or F-type stars. An IR excess or Br γ emission that would be readily apparent for a late-type star may be overwhelmed by the greater photospheric flux from these stars. Thus, these apparently non-accreting stars may still be accreting some matter. All targets, including the apparent non-accretors, were selected to have increasing flux with wavelength as observed by IRAS, so they all have significant circumstellar material. Non-accretors (excluding targets where there was another YSO in the IRAS beam) also tend to have a higher median spectral index than the sample as a whole (1.3 versus 0.5 from λ= 12 μm to 100 μm), suggesting that the non-accretors tend to have relatively less warm dust and more cool dust than the accretors, consistent with the scenario of a cleared disk hole.

We expected that the accretion luminosity (L_bolometric − L_star) would be proportional to the veiling. Accretion luminosity could only be estimated for those stars for which we have estimated the photospheric spectral type. Analysis of the bolometric luminosity (determined from IRAS observations) or the accretion luminosity showed no trend with our derived values of the veiling (r_k). There are several objects with a late-type spectral classification, very high accretion luminosities, and negligible veiling. For many of these cases, there are other deeply embedded YSOs within the broad IRAS beam. To show how the veiling is related to the accretion luminosity, high angular resolution mid- and far-IR data are needed to spatially resolve these YSOs and higher spectral resolution near-IR spectra may be needed to more accurately determine the photospheric spectral types.

Several targets have spectra that cannot be well modeled by adding extinction and veiling to a stellar photosphere. The most common of these are targets with FU Orionis-like spectra, characterized by having a near-IR spectrum dominated by water vapor absorption, without clearly defined narrow photospheric absorption lines, few if any emission lines, and deep CO absorption in excess of what is observed in the spectra of late-type dwarf photospheres. Flux from these objects is dominated by emission from the disk (presumably due to accretion luminosity) with negligible flux from the stellar photosphere (Hartmann & Kenyon 1996). Since the heating of the disk is dominated by viscous dissipation versus stellar irradiation, the disk interior is hotter than the surface. Cooler material above the disk photosphere imprints the observed water and CO absorption bands on the spectrum. One object (04073+3800) has a highly veiled spectrum without excess water absorption and has been identified as an FU Orionis-like star (Sandell & Aspin 1998). FU Orionis-like objects are found in region E of Figures 4 and 5. As a group, these targets are much more luminous than the rest of the observed sample. The median bolometric luminosity for targets with FU Orionis-like spectra is 28 L_☉, whereas the median bolometric luminosity for targets in region D (with dwarf-like spectra and gravities) of Figures 4 and 5 is 4.0 L_☉. The two-sample K-S test shows that those targets with FU Orionis-like spectra have a higher bolometric luminosity than those targets with dwarf-like spectra with 99.5% confidence. Although the bolometric luminosities are higher, the spectral index distributions (both of which are measured using IRAS fluxes from 12 μm to 100 μm) for regions D and E are statistically indistinguishable. Since these targets have deep CO absorption and tend to be excessively luminous, these are likely to be examples of FU Orionis-like objects. Indeed, two targets in our sample with this kind of spectrum, IRAS 04287+1801 (L1551 IRS5) and IRAS 21454+4718 (V1735 Cygni), are well-known FU Orionis-like objects.

Of the 10 FU Ori and FU Ori-like stars in our sample, two (and possibly a third) are newly identified FU Ori-like candidates and these are the first near-IR spectra of them. FU Orionis-like objects are relatively rare. Previously, ∼21 have been identified (Reipurth & Aspin 2010) and only 9.1% ± 2.9% (10/110) of the sample targets have this type of spectrum. Thus, the fraction of Class I YSOs that pass through an FU Orionis phase times the average duty cycle of the FU Orionis phase is roughly 9%. It is significant to find such a large fraction of FU Orionis-like stars among a sample of Class I YSOs. Identifying two new embedded FU Orionis-like stars is particularly important since these observations have significantly increased the number of known embedded FU Orionis-like objects. Previously, only ∼8 have been identified. Table 6 lists the FU Ori and FU Ori-like stars in our sample, as well as other stars that show excess CO absorption.

In addition to FU Orionis-like stars, seven other stars show CO absorption greater than what would be expected from the star's spectral type. All of these show photospheric absorption lines, and in four cases the CO EW is only marginally greater than the M4 value and we do not consider these four to be candidate FU Orionis-like objects. IRAS 06393+0913 has strong CO absorption but lacks the high veiling and deep water absorption bands that are commonly observed in the spectra of FU Orionis-like objects, and thus we do not classify it as an FU Orionis-like object. Three targets (IRAS 05327−0457W, 05404−0948, and 16235−2416) have an A- or F-type photosphere with weak CO band-head absorption. IRAS 05404−0948 also appears to have weak water absorption bands. In these cases, the CO band-head and water absorption may be caused by circumstellar material or an unresolved late-type companion. The absolute K-band magnitudes (simply the observed K-band magnitude of these objects corrected for distance, and not corrected for accretion or extinction) of the K- and M-type stars in this sample are on average 2.5 mag fainter than the absolute K-band magnitudes of the A- and F-type stars (in comparison, the bolometric luminosities of main-sequence A0 and M0 stars are different by ∼8 mag). However, the brightest K- and M-type stars have the same absolute K-band magnitude as the faintest A- and F-type stars, most likely due to a combination of accretion and extinction-related effects. As such, for Class I YSOs it is reasonable to expect to see a contribution from a late-type companion star in the spectrum of an early-type primary star. The other three stars (04292+2422W, 04591−0856, 18341−0113N) are best fit with G- or K-type spectra but have CO absorption in excess of the latest type star that is a good fit. Remarkably, these six objects also tend to not have emission lines. IRAS 04591−0856 has H₂ emission lines, which may be associated with a wind (Beck et al. 2008), and 05327−0457W shows weak Br γ emission at the bottom of that deep absorption feature. Considering that these six objects share some properties in common with FU Ori-like stars (excess CO absorption, a dearth of emission features), these objects may be experiencing weak FU Orionis-like activity. Thus, the shape of the continuum and the depth of the CO band heads in the spectrum of a Class I protostar may not always be representative of the true spectral type of the stellar photosphere.

Herbig Ae/Be stars (Herbig 1960) were identified as stars with spectral type earlier than F0 with Balmer emission lines and associated with a dark cloud. Hillenbrand et al. (1992) showed that these are intermediate-mass pre-main-sequence stars with massive accretion disks. This spectroscopic survey of Class I YSOs found three targets with an A class best-fit spectral type (see Table 2), as determined by our spectral line matching routine. Since these objects satisfied our criteria to be Class I YSOs, and since the spectra of two of these stars show emission features, all three show veiling, we believe that these are embedded Herbig Ae stars. In this case, only 2.7% ± 1.6% of the Class I YSOs are embedded Ae-type stars.

Many T Tauri binary systems have been spectroscopically observed, and Table 1 in Monin et al. (2007) lists the number of T Tauri binary systems with classical T Tauri (possessing a disk and accreting matter) and weak line T Tauri (non-accreting) components. We used the number of T Tauri binaries in the center column of Table 1 in Monin et al. (2007) to exclude objects with very close companions, and excluded the three systems with a passive disk. Table 1 in Monin et al. (2007) includes 59 (74.6% ± 9.7%) T Tauri binary systems where both components are classical T Tauri stars or both are weak line T Tauri stars, whereas 20 (25.3% ± 5.7%) are mixed systems. Has this correlation been established by the Class I phase?

Over the course of this survey, we observed the spatially resolved spectra of both of the components of nine previously identified multiple protostars included in Connelley et al. (2008a). This includes 1 triple system for a total of 10 pairs. In order to be clearly resolved, these objects are well separated on the sky, with a median angular separation of 45. These objects have a median projected separation of 2700 AU, with a minimum separation of 400 AU and maximum separation of 4500 AU. Both components of each pair also had to be brighter than K = 12 to get a high S/N observation. The properties of these binaries are given in Table 7.

We noticed that the spectra of Class I binaries tend to have the same veiling, i.e., both components have high veiling or both have low veiling. Are the veiling of the binary components consistent with randomly pairing spectra from the whole sample? In the whole sample, 57.4% of targets have high veiling, whereas 42.6% have low veiling. Among the binary spectra, both components of 8/10 pairs (80% ± 28%) either both have high veiling or both have low veiling. The probability that less than 3 out of 10 pairs would have different veilings, having been randomly selected from this atlas of protostellar spectra, is 4.6%. Thus, we can state at the ∼95% confidence level that binary protostars are not randomly paired and that their spectra tend to have the same veiling. Excluding the FS Tau system, which includes a well-known T Tauri star, both components of 8/9 pairs (89% ± 31%) either both have high veiling or both have low veiling. Subjectively, the spectra of binary components often look remarkably similar. For example, the spectra of the components of IRAS 18383+0059 and the components of IRAS 05375−0040 are quite difficult to tell apart. We have shown that the trend of T Tauri binary components to both be accreting (or not) extends to the Class I phase at much higher accretion rates, for targets selected across the sky. We note that 8/11 of the components with low veiling show Br γ emission, and thus are accreting.

Interactions between the circumstellar disks could explain the dearth of mixed Class I binary systems. For example, a close passage between two stars could trigger higher accretion and/or higher veiling in both systems. The projected separations between the stars of these binaries are within an order of magnitude of the expected size of a circumstellar disk, so interactions between the disks are plausible. If this scenario is true, then we would expect that closer binaries would tend to have higher veiling. Based on the values in Table 7, there is no trend of increasing veiling with decreasing projected separation.

If interactions between the disks are not responsible for these stars having the same veiling, and if we assume that these protostellar binaries are coeval, then another explanation is that the veiling systematically changes with time within the Class I phase. Presumably, younger objects have higher veiling and more evolved objects have lower veiling. If the veiling of a given protostar were not to systematically change with time, but rather changed randomly on a timescale much less than the Class I life time, then the number of binary pairs with the same (or different) veilings would be indistinguishable from a random selection from the whole sample.

H₂ emission is seen from 42.7% of the targets in the whole sample. Similarly, Doppmann et al. (2005) found that 44% of Class I and flat spectrum YSOs have H₂ emission in their smaller sample. Among the binary spectra, both components of 9/10 pairs (90% ± 30%) either have H₂ emission or both do not have H₂ emission. The probability that only 1 out of 10 pairs would have different H₂ emission, having been randomly selected from this atlas of protostellar spectra, is 2.8%. The presence of H₂ emission in both components of protostellar binaries is correlated with ∼97% confidence. We also note that only 5 of 19 binary components (26.3%) have H₂ emission less than the 42.7% for the whole sample.

Lucas et al. (2001) found that a number of young (∼1 Myr) low-mass (<0.1 M_☉) sources in the Trapezium cluster have an H-band continuum spectrum that is characterized by a triangular profile with a peak near 1.68 μm. They attribute this to the expected lower gravity of younger sources, and that this is an independent way to verify that these are substellar sources. A field dwarf observed by Kirkpatrick et al. (2006) also has a triangular-shaped H-band continuum spectrum which they interpret as a sign of low gravity and thus youth since regular field dwarfs (Cushing et al. 2005) do not exhibit this feature. Several other young brown dwarfs have been identified (e.g., Gizis 2002; Looper et al. 2007) with this feature. The surface gravities of protostars and young brown dwarfs are very similar. The average log(g) for protostars from Table 3 of Doppmann et al. (2005) is 3.98, whereas average log(g) for 5 Myr brown dwarfs from Table 4 of Rice et al. (2010) is 3.87. The triangular shape of the H-band continuum spectrum is caused by the reduction of the pressure-induced opacity in H₂ in the atmosphere of the star (M. Marley et al. 2010, in preparation). The spectrum of pressure-induced opacity in H₂ is very broad, which has the effect of smoothing the opacity spectrum of the stellar atmosphere as a whole. Without this opacity source, the shape of the continuum near the H band is dominated by the opacity spectrum of water.

We used our continuum fitting routine described above to determine which objects in our sample with a late-type continuum (excluding FU Orionis-like objects) also have a triangular H-band continuum. These objects are flagged in Table 1. As an input to this routine, we used the best-fit spectral type as well as the latest spectral type within the uncertainties. Of the 27 objects in our sample with a well-matched late spectral type, 15 (55.5% ± 14.3%) have a triangular H-band continuum. Of the 21 objects in our sample of the M class spectral type, 10 (47.6% ± 15.1%) have a triangular H-band continuum. These results show that the triangular H band is common among Class I YSOs with late-type spectra, yet only about half of them show this feature. Among all Class I YSOs, only about 14% of them have a triangular H band. Although our sample of objects are certainly all very young (indeed much younger than the brown dwarfs discussed above), they are likely to be much more massive than brown dwarfs. The objects in our sample may have evolved enough (toward higher gravities) so that only about half of the late-type spectra in the sample show this triangular H-band structure. Also, high extinction and high veiling make it more difficult to identify which targets have a triangular-shaped H-band continuum, so this result should be interpreted as a lower limit.

Greene & Lada (1996) used the EW of Na i (2.206 μm and 2.208 μm) plus Ca i (2.263 μm and 2.266 μm) versus the EW of CO as a gravity-sensitive indicator. On this plot, most YSOs fall closer to the line formed by luminosity Class V than luminosity Class III stars, with FU Orionis-like stars below the line made by luminosity Class III stars. Figures 4 and 5 show the locations of our Class I protostars on this plot. The EWs for the luminosity Class III (dwarf) and V (giant) stars were measured from stars in the SpeX spectral library in the same way as our YSOs. Similar to the result in Greene & Lada (1996), most of our Class I YSOs also lie closer to the line formed by luminosity Class V than luminosity Class III stars (region D), with increasing veiling pushing the location of stars on this plot toward (0,0). Class I YSOs tend to have slightly lower gravity than dwarf stars, but not as low as giants. YSOs with triangular H-band continua are shown in red in Figures 4 and 5, and tend to be found to the upper right of the remaining YSOs.