THE PHOTOEVAPORATIVE WIND FROM THE DISK OF TW Hya^*

Because one of our goals is to measure small shifts in the peak of the [Ne ii] line, we performed additional tests to verify the instrument wavelength solution and investigate the relative and absolute wavelength accuracy of VISIR in the high-resolution long-slit mode. First of all, we reduced and analyzed eight additional archival spectra from three different infrared-bright K4-6 III standard stars (HD 136422, HD 139127, HD 149447) observed in the same setting as TW Hya. After applying the proper heliocentric correction and shifting the spectra in the stellocentric frame, we compared the observed peak position of two photospheric absorption lines to those predicted by the MARCS model atmosphere of a K5 III star (HR 6705; the spectrum was kindly provided by L. Decin). While differences in velocities are confined to within ±2 km s⁻¹ it is apparent that one of the lines is typically redshifted while the other is typically blueshifted. To investigate whether there could be an issue in the wavelength dispersion solution and whether errors could be larger at other wavelengths, we turned to spectra of Titan which were acquired with the same setting as TW Hya (VLT/VISIR PID 083.C-0883). Titan spectra present many unresolved lines from C₂H₂ and C₂H₆ molecules in this setting. In a case of large drifts in wavelength, we might see the FWHM of unresolved lines change across the spectrum. We measured the FWHMs of seven unresolved emission lines in two spectra of Titan covering the full VISIR wavelength range (three before and four after the [Ne ii] line in TW Hya) and found no such trend; rather we report a narrow range of FWHMs with a mean value of 10 km s⁻¹, the spectral resolution of VISIR in this setting. In addition, we compared the location of Titan lines with the rest wavelengths reported in the HITRAN database (Rothman et al. 2009). In the setting covered with VISIR there are seven transitions from the C₂H₆ molecule, two of which are single and can be used to measure the relative centroids to other lines through the spectrum. Figure 3 shows that the relative uncertainty between the lines is 1 km s⁻¹, one-tenth of the spectral resolution of VISIR in this mode. However, as hinted by the test on the standard stars and as we will see on the TW Hya data (Section 4), the absolute accuracy reachable with VISIR is likely a factor of two higher (2 km s⁻¹), due to uncertainties in placing the source in the center of the slit.

Zoom In Zoom Out Reset image size

Spitzer/Infrared Spectrograph (IRS) spectra of TW Hya taken in three different years show variations of up to ∼30% in the continuum emission, as well as in the [Ne ii] flux and equivalent width (Najita et al. 2010). These differences may arise from pointing errors and/or true source variability. The VISIR spectra we acquired in this observational campaign provide additional epochs to further investigate continuum and line variability on the timescale of years as well as on the much shorter timescale of hours.

For each VISIR spectrum we compute the location of [Ne ii] peak emission in the stellocentric frame (after assuming a stellar radial velocity of 12.2 ± 0.5 km s⁻¹; Weintraub et al. 2000), the equivalent width (EW), and when standard stars were acquired close in time to the TW Hya spectroscopic exposures, as well as the line flux and continuum near the line (essentially for each P.A. in the first night and for the first P.A. in the second night). Peak positions and [Ne ii] line fluxes are calculated from Gaussian fits to the observed line profiles while EWs are computed by integrating the line flux within ±4σ of the fitted Gaussian. Table 3 summarizes the main results and Figure 2 shows line profiles for the five flux calibrated spectra.

To test the hypothesis of variations due to calibration uncertainties, we search for correlations between the main properties of the [Ne ii] line and the adjacent continuum with the airmass and the location of the source in the slit. We find an anti-correlation between the line flux and continuum and the source airmass (lower airmass→higher flux and continuum, see Figure 4) suggesting that most variations in these quantities are due to flux calibration uncertainties. We do not find any correlation between these quantities and the location of the source in the slit. However, we note that the displacement of the source with respect to the center of the slit might be the cause for the lowest EW and the largest blueshift in the [Ne ii] line for P.A. = 63° (see Figures 1 and 4). Unfortunately, we do not have a standard star to flux calibrate the TW Hya spectrum at this P.A. but our expectation is a reduced line flux and continuum. We also note that the highest [Ne ii] flux we measure (for P.A. = 180°, smallest airmass) is as high as that measured by Spitzer/IRS within the estimated error bars. This demonstrates that the small VISIR slit recovers all the [Ne ii] emission from the disk of TW Hya implying that the emission is confined within ±10 AU from the star.

Zoom In Zoom Out Reset image size

[Ne ii] line EWs are less affected by differences in the airmass: for the five exposures at airmass <1.3 (excluding P.A. = 63°) the mean EW is 46 ± 2 Å. This value is slightly larger than the [Ne ii] EW from our 2008 high-resolution spectrum (Pascucci & Sterzik 2009), the same as the 2006 EW measured with Spitzer/IRS (Najita et al. 2010), and smaller by about 20 Å than the 2004 and 2008 EWs measured with Spitzer/IRS and the 2007 EW measured with Gemini/Michelle by Herczeg et al. (2007). Although the sampling is sparse, the [Ne ii] EW essentially oscillates around two values (∼45 and 65 Å) which suggests intrinsic variability in the [Ne ii] emission or nearby continuum on a timescale of ∼1–2 years.

We also confirm the blueshift in the [Ne ii] line with respect to the radial velocity of TW Hya (see Figure 4 second panel). The mean blueshift is −5 ± 1 km s⁻¹ in agreement with our previously reported blueshift (Pascucci & Sterzik 2009). The peak position of the [Ne ii] line is not correlated with the airmass at which the source was observed, nor with the location of the source in the slit (but see previous note for P.A. = 63°). The FWHM of the [Ne ii] line presents the least variations with P.A., and is not correlated with the EW, peak position, or the airmass: the mean value and standard deviation for the FWHM are 16.3 ± 0.9 km s⁻¹ or an intrinsic FWHM of 12.9 ± 0.9 km s⁻¹ after deconvolution with the instrumental FWHM of VISIR of 10 km s⁻¹.

We search the FEROS optical spectra for all the lines predicted by Font et al. (2004), Hollenbach & Gorti (2009), and Ercolano & Owen (2010) to have luminosities of a few 10⁻⁷ L_☉ or higher, thus comprising forbidden and semi-permitted transitions from O i, O ii, Mg i, N ii, and S ii. We report equivalent widths and upper limits in Table 4 for representative transitions. To search for weak lines and properly measure EWs we had to correct for the numerous photospheric absorption lines present in the optical spectra of TW Hya. We followed the approach described by Alencar & Batalha (2002) and divided the continuum normalized spectra of TW Hya with the artificially veiled and continuum normalized spectrum of GJ 1172, a K7 V star with similar v sini as TW Hya (see also Hartigan et al. 1989 for the procedure and definition of veiling). In addition, we combine all observed spectra in a median high signal-to-noise spectrum normalized to the continuum level.

Optical spectra from T Tauri stars are known to have strong forbidden emission lines from [O i] at 6300, 6364, and 5577 Å (Hartigan et al. 1995). TW Hya is no exception and these transitions are easily identified in all individual spectra. The strongest of all is the [O i] line at 6300 Å with a median EW of 0.47 Å, which was found not to vary significantly with time (Alencar & Batalha 2002). Following Hartigan et al. (1995) and assuming a mean R magnitude of 10.2, we convert this EW into a line luminosity of 10⁻⁵ L_☉. The optical spectrum of TW Hya is known to vary, the most detailed V magnitude monitoring reports night-to-night variations of ΔV ≃ 0.35 (seasonal changes have smaller amplitudes; Rucinski et al. 2008). The same variations in R convert into a [O i] luminosity range of (0.8–1.6) × 10⁻⁵ L_☉. The 6364 Å [O i] line is a factor of ∼3 weaker than the [O i] 6300 Å, as expected from the ratio of the Einstein coefficients, while the 5577 Å [O i] line is a factor of ∼7 weaker on average (see Table 4). These lines are also spectrally resolved with FEROS, the mean (and median) FWHM of the [O i] 6300 Å is 11.5 ± 0.4 km s⁻¹ or an intrinsic line width of ∼10 km s⁻¹ after deconvolution with the 6 km s⁻¹ instrumental FWHM. The other weaker [O i] lines have similar FWHMs. As already reported in Alencar & Batalha (2002) the [O i] 6300 Å line presents only a narrow component, is centered at the stellar rest frame, and there is no indication of a blueshifted high-velocity component from jets/outflows. We use 40 spectra from Alencar & Batalha (2002) to further investigate whether there is a blueshift of just a few km s⁻¹ in the [O i] peak emission.¹² To do that we rely not only on the absolute wavelength calibration performed previously but also compare the location of the [O i] 6300 Å line with respect to other known photospheric lines. Figure 5 shows the distribution of velocities for the [O i] 6300 Å line in the stellocentric frame measured via the photospheric Ca i 6439 Å line (black dashed line). This histogram peaks at negative velocities but the median difference velocity is only −0.1 km s⁻¹ with a standard deviation as large as 0.15 km s⁻¹. While this tiny blueshift might indicate a flow, we show in the same figure that it is almost certainly related to small differences in the dispersion solution between the two different echelle orders in which the [O i] 6300 Å and the Ca i 6363 Å fall. To demonstrate this we overplot the difference between the [O i] 6300 Å and the [O i] 6364 Å which falls in the same order as the Ca i photospheric line and should trace the same disk emission as the [O i] 6300 Å line. Figure 5 clearly shows that the two velocity distributions are identical, the median difference between the [O i] 6300 Å and the [O i] 6364 Å being −0.1 km s⁻¹ and standard deviation being 0.16 km s⁻¹. These tests demonstrate that the [O i] optical lines are not blueshifted.

Zoom In Zoom Out Reset image size

In addition to the oxygen lines, we also detect the [S ii] line at 4069 Å and the Mg i] line at 4571 Å with luminosities that are about ∼10 times lower than the [O i] 6300 Å luminosity (Table 4). The [S ii] transition at 4076 Å comes from very similar upper energies (∼3 eV) as the 4069 Å transition but the Einstein coefficient is 2.5 times smaller, consistent with our non-detection and 3σ upper limit. The [S ii] at 4069 Å and the Mg i] line at 4571 Å peak at the stellar radial velocity as the [O i] lines. Their luminosities are similar to those predicted by the X-EUV primordial disk model with log(Lx) = 29.3 (Ercolano & Owen 2010). However, we note that this X-ray luminosity is low for TW Hya, about an order of magnitude lower than what has been estimated by, e.g., Kastner et al. (1999). In the framework of photoevaporative disk models, the most puzzling result is the non-detection of the other two [S ii] transitions at 6716 and 6731 Å with stringent upper limits pointing to fluxes ∼50 times lower than the [O i] 6300 Å line. As mentioned in Section 2, both EUV and X-ray irradiated disks should produce a strong [S ii] doublet for gas at low density and flowing, similar in strength or just a factor of a few weaker than the [O i] 6300 Å line. While the [SII] 6716/6731 doublet should trace more extended low density gas than other forbidden lines due to its much lower critical density,¹³ it is unlikely that the FEROS spectra lost most of the [SII] 6716/6731 flux since the fiber diameter projected on the sky is 27 and covers radii out to ∼70 AU at the distance of TW Hya. Thus, the non-detection of these transitions points to specific conditions in the disk emitting region, specifically to dense gas, as we shall discuss later.

Finally, the centrally peaked profiles and the FWHM of the optical lines can help constrain the extension of the emitting region. The [O i] 6300 Å line has an intrinsic width of 10 km s⁻¹, the [SII] and MgI] lines are similarly narrow ∼10–12 km s⁻¹, hence likely trace similar disk radii. For a given ring of gas in Keplerian rotation extending out to a radius r_o from the star the absence of a double peak implies $2\,\sqrt{(\frac{GM_{\star }}{r{\rm o}})} \sin (i) < \Delta v_{\rm FEROS}$ or r_o > 0.4–1 AU for the mass and disk inclinations reported for TW Hya (Table 2). Again for gas in Keplerian rotation, the inner ring radius r_i determines the maximum velocity in the line profile. Because the observed [O i] 6300 Å profiles extend out to ±10 km s⁻¹ (see Section 4.3) the inner radius of the gas ring should be <0.2 AU and can be as close to the star as 0.06 AU for a 4° disk inclination. This inner disk gas is likely not photoevaporating. However, if most of the emission is extended beyond about 1 AU and gas is photoevaporating, detailed disk models are necessary to compute line profiles and estimate the extension of the emitting region (see Section 5.2).

Our first high-resolution VISIR spectrum of TW Hya showed a hint for an asymmetry in the [Ne ii] line profile (see Figure 4 from Pascucci & Sterzik 2009), with possibly more emission on the blue side than on the red side of the line. A flux enhancement on the blue side was also pointed out by Herczeg et al. (2007) in their lower signal-to-noise spectrum. Such an enhancement is expected for face-on disks with an optically thick midplane blocking the view of the redshifted outflowing gas (Alexander 2008; Ercolano & Owen 2010). We use the eight additional spectroscopic exposures acquired in this observational campaign to further investigate the [Ne ii] 12.81 μm profile. To check for asymmetries we apply the following method. For each line we produce a mirror spectrum (in velocity scale), then we apply small shifts in velocities to the mirror spectrum such that the upper half of the lines matches best, and finally we subtract the shifted mirror and original spectra. This method is especially sensitive to detecting any blue or red excess emission in the wings of the line compared to its core emission and is more suited to lines that are poorly sampled in velocity than the line bisector method (e.g., Gray 1980).

Figure 6 summarizes our findings. Individual line profiles as well as the median of line profiles (thick black line) show an excess emission on the blue side of the [Ne ii] line peaking around −20 km s⁻¹ and a typical shift of the peak centroid of −5 km s⁻¹.¹⁴ The blue excess is also visible as decreased emission on the red side of the line at ∼10 km s⁻¹ (see upper and lower panels of Figure 6). We can confirm that this asymmetry is intrinsic to the source emission and not an instrumental effect by applying the same method to spectrally unresolved emission lines from Titan (see Figure 7).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We then apply the same method of mirrored spectra to simulated [Ne ii] line profiles from photoevaporative disk winds. As explained in Section 2, such profiles are available for the EUV-only (Alexander 2008) and the X-EUV (Ercolano & Owen 2010) models. These models use input parameters such as stellar mass, X-ray, and EUV luminosities that are similar to measured values for the TW Hya system but were not specifically tuned for TW Hya (in the case of the X-EUV we use the models with log(L_X) = 30.3 which are appropriate for TW Hya). In all cases the disk midplane is optically thick thus introducing an asymmetry in the line profile which is, as expected, more pronounced the closer the disk is to the face-on orientation (see Figure 8). Applying the K-S test to the difference curves shows that the XEUV 10° disk inclination model has just a 6% probability that the observed and model profiles are drawn from the same parent population hinting at a difference. All other models (EUV with i = 3°, 10° and XEUV with i = 3°) have probabilities as high as 40% meaning that they are all equally good matches to the data. The important result from this analysis is that the [Ne ii] emission must arise from a region in the disk that is optically thick to the redshifted outflowing gas. We shall discuss in Section 5 what this implies for the extension of the ionized wind component of the photoevaporative flow.

Zoom In Zoom Out Reset image size

Finally, we also apply the method of mirrored spectra to the strongest of the optical lines, the [O i] 6300 Å line, to check for asymmetries in the line profile. Figure 9 shows that, unlike the [Ne ii] line, the [O i] profile is symmetric and the line is centered at stellar velocity. As we will discuss in more detail in Section 5, this profile could result from (1) pure Keplerian motion of gas bound to the disk (i.e., the [O i] is not tracing the neutral component of the wind, it is not flowing) or (2) a photoevaporative flow from a disk region that is optically thin to the redshifted outflowing gas. Examples of symmetric lines from photoevaporative flows are presented in Ercolano & Owen (2010) for disks with inner holes in their dust and gas distribution and emission confined within the hole. Because most of the opacity is expected to be from dust (see Section 5) and dust extinction decreases with wavelength, our line profile analysis demonstrates that if scenario (2) applies the neutral [O i] gas does not trace the same radii as the [Ne ii] emitting gas.

Zoom In Zoom Out Reset image size

We attempted for the first time spectroastrometry in the mid-infrared to improve upon the spatial resolution achievable with the VLT/VISIR and resolve the innermost region of the photoevaporative wind. Spectroastrometry relies on the fact that the measurement of the centroid of a flux distribution can be determined to much better than a pixel precision depending on the signal-to-noise of the observations. In the case of VISIR one pixel corresponds to 127 mas meaning that we could investigate the [Ne ii] emission at scales better than 6 AU at the distance of TW Hya. Asymmetries in the emission at a specific wavelength appear as shifts in the centroid of the flux distribution. Using different slit P.A.s enables us to determine the orientations at which asymmetries are stronger and thus helps constrain the distribution of the source emission.

To retrieve the spectroastrometric signal from our VLT/VISIR data we extract spatial profiles in each velocity channel and measure the spatial centroid of the [Ne ii] emission in cross-dispersion direction with respect to the centroid of the emission in the continuum. The continuum is measured as the mean of the continuum centroids on the red and blue sides of the [Ne ii] line. Because of the low S/N on the continuum we summed up the spatial profiles over more than 10 pixels in velocity; we have tested that our results do not vary if we change the size of the velocity bins or the location where we measure the continuum. To determine the centroid of the spatial profiles we have tested Gaussian and Lorentzian profiles as well as the weighted mean over the FWHM of VISIR in spatial direction as proposed by Pontoppidan et al. (2008). We find that fitting the emission with Lorentzian and Gaussian profiles provide the same spectroastrometric signal and a more robust centroid than the weighted mean approach. As uncertainty on the centroid we take the inverse of the S/N for each spatial profile where the noise is measured as the standard deviation of the continuum outside the profile. This error reflects that profiles extracted at velocities sampling the wings of the line have lower S/N, hence the centroid is less well determined. The centroid is also diluted by the continuum flux at each velocity to take into account the different continuum-to-line ratio moving from the core to the wing of the line (e.g., Whelan & Garcia 2008). Figure 10 shows the centroids in pixels versus velocities in the stellocentric frame for the parallel and antiparallel slit orientations separately. This way we can better search for any possible instrumental effects on the centroids because such effects would appear as shifts of the same sign and intensity in the couple of parallel and antiparallel slits (Bailey 1998). The plot does not show any strong similar effect but small amplitude modulations across the [Ne ii] line and nearby continuum. As we will see next, the centroids are measured to an accuracy of a tenth of a pixel close to the [Ne ii] peak emission and typically to half of a pixel in the wings where the S/N of the flux distribution in cross-dispersion direction is low.¹⁵

Zoom In Zoom Out Reset image size

4.4.1. Comparison between Observations and EUV-model Predictions

The simple EUV-only disk model is computationally fast enough that we can create channel maps for the [Ne ii] emission and compute spectroastrometric curves for the slit orientations used in this observational campaign. The model is essentially identical to the EUV-only model used by Font et al. (2004) and Alexander (2008) with a radially continuous gas distribution. The only differences between our hydrodynamic model and that of Alexander (2008) are the resolution of the model grid and the location of the lower boundary. As we wish to generate spatially resolved emission maps, it is necessary to use higher spatial resolution than that used by Alexander (2008). In addition, we now assume that the underlying disk has a finite thickness and place the base of the ionized wind at z/r ≃ 0.15 (e.g., Hollenbach et al. 1994). This has only a small effect on the predicted line profile, reducing the peak blueshift by approximately 1 km s⁻¹. Our polar [(r, θ)] computational grid covers the range r = [0.03r_g, 10.0r_g], θ = [0, 75°], with grid cells that are logarithmically spaced in r and linearly spaced in θ. We use N_r = 1113 cells in the radial direction and N_θ = 250 cells in the polar direction. When the grid is expanded to three dimensions (by assuming reflective symmetry about the disk mid-plane and azimuthal symmetry around the polar axis), the resulting grid has 1113 × 600 × 1200 cells in the r, θ, and ϕ directions, respectively.

We compute the line emission from each grid cell in the same manner as Alexander (2008). However, following the results of Hollenbach & Gorti (2009) we adopt an ionization fraction for Ne ii of 0.75 (rather than the somewhat arbitrary value of 0.33 used previously). For comparison to the TW Hya system we adopt a stellar mass of M_* = 0.7 M_☉ (thus r_g = 6.2 AU) and a disk inclination of i = 4° (see Table 2 for references). We set the ionizing flux Φ by matching the predicted line luminosity to the observed value of 3.1 × 10⁻⁶ L_☉: we find that Φ = 7.5 × 10⁴⁰ ionizing photons per second matches the observed line flux to ∼1% accuracy. The integrated mass-loss rate from the disk is ∼1.5 × 10⁻¹⁰  M_☉. We create an integrated line profile (identical to that of Alexander 2008) by summing the contributions from every cell on the grid, and also generate spatially resolved emission maps by de-projecting the three-dimensional grid on to the sky plane. This is done at each (spectral) velocity to generate a three-dimensional data cube. The model data cube has 201 × 201 pixels which are 0.255 AU on a side (equivalent to 5 mas at the 51 pc distance of TW Hya) and a velocity resolution of 0.1 km s⁻¹. We model the spectral resolution of VISIR by convolving the data cube (in the velocity direction) with a Gaussian profile with half-width σ = 5 km s⁻¹, to give a simulated spectral resolving power of R = 30, 000.

Figure 11 shows two simulated images of the [Ne ii] emission taken at a negative velocity and a positive velocity; the image orientation is as in Hughes et al. (2010) when N is up and left is E in the figure. To simulate the spectroastrometric signal we apply a slit in the x-axis equal to the VISIR slit width of 400 mas, collapse the flux in the x-axis, and evaluate the centroid of the spatial profile in the y-axis after convolution with the VISIR point-spread function (PSF; FWHM of 300 mas), for each velocity channel. To simulate the rotation of the slit we rotate the simulated images by the corresponding angle and then apply the slit as described above. Figure 12 shows the expected spectroastrometric signal as parallel minus antiparallel centroids in VISIR pixels for the photoevaporative disk model (solid lines) and for the wind only, i.e., when Keplerian rotation is set artificially to zero (dashed lines). One sees that for a slit orientation close to the disk major axis (P.A. = 333°) the spectroastrometric signal is dominated by the Keplerian component of the gas, the asymmetry in the emission is small but present due to the small but non-zero disk inclination. The positive displacement at negative velocities can be understood by looking at Figure 11: the emission at negative velocities is stronger in the direction of P.A. = 333° and slit orientations close to it. The wind component dominates the signal for the slit orientation perpendicular to the disk major axis (63° but it is prominent also for 270°) because at that orientation the Keplerian component is close to zero. The main results from the modeling are as follows: (1) the expected shift in centroid is small, about 0.2 pixels, and (2) the shift is primarily set by projection effects (disk inclination) and the Keplerian component of the gas since the wind is more symmetrically distributed around the star.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

To compare model predictions with observations, we subtract the measured antiparallel centroid positions from the parallel positions (after interpolation over the same velocity grid), see Figure 13. For P.A.s 270°and 63°, slits oriented close to the disk minor axis, the spectroastrometric shift is less than 0.1 pixel as expected. For the other two slit orientations, sampling the disk major axis, we do not detect the S-shape profile predicted for gas in Keplerian rotation. We see a hint of such a profile only for P.A. = 333° negative velocities which is, however, not present at the positive velocities. A smaller shift than predicted could result from a [Ne ii] emission that is more symmetric than predicted, from a smaller disk inclination, and/or from [Ne ii] emission more closely concentrated toward the central star (in this last case, the convolution with the VISIR PSF would smear out the contrast between the bright and faint rims of the [Ne ii] emission). Further observations at even higher spectral resolution and sensitivity, especially in the wings of the line, are necessary to confirm these first results and to possibly identify the small spectroastrometric shift from photoevaporating gas that is expected for slit orientations along the minor axis.

Zoom In Zoom Out Reset image size

The small blueshift in the [Ne ii] line is a clear tell-tale sign of photoevaporating gas for a face-on configuration like that of TW Hya. Hence, we are sure that this line is probing the ionized component of the disk wind. This component could be the dominant wind component in the EUV case or a trace component of the mostly neutral wind in the X-ray case. Our observations constrain the radial extension of the ionized wind component via the [Ne ii] line. By recovering all flux measured with Spitzer/IRS, we demonstrate that the [Ne ii] emission arises within 10 AU of the star. We note that although the radial extension of the mass loss is very different for pure EUV and X-EUV irradiation, with X-rays producing 50% of the mass loss at tens of AU (Owen et al. 2010), the [Ne ii] luminosity decreases steeply with radial distance from the star (as ∼1/r) hence the emission is dominated by the flux from small radii. For the specific EUV and X-EUV models discussed here 90% of the [Ne ii] emission arises within 12 AU and 15 AU, respectively, which is consistent with the 10 AU radial extension inferred observationally.

The blue excess in the wing of the [Ne ii] line adds a further constraint on the radii traced by this wind component. We have shown in Section 4.3 that primordial disks (no gaps) irradiated by EUV and X-rays can reproduce the [Ne ii] line profile and peak emission. Here, we ask the question: how much [Ne ii] emission could originate in an optically thin region and still be consistent with the observed [Ne ii] profile? To simulate the [Ne ii] profile from optically thin disk radii we take a Gaussian centered at zero velocity and FWHM set by thermal broadening for gas at ∼5000 K (X-ray heating) or ∼10,000 K (EUV heating) and convolved with the FWHM of the VISIR spectrograph. We add as much as 10%, 20%, and 30% of the total emission to the [Ne ii] profile from a fully thick disk and compare the resulting profile with the observed one. We find that already a 10% contribution from radii optically thin to the [Ne ii] emission shifts the line peak emission redward to ∼−4 km s⁻¹, marginally consistent with our measurements, and increases the symmetry of the line. This simple approach shows that more than 80% of the [Ne ii] emission should come from a disk region that is fully optically thick to the [Ne ii] line. What is then the source of opacity within 10 AU from the central star? This is a relevant question for transition-type sources like TW Hya whose SEDs clearly point to opacity gaps in their disks.

First we look at the gas component and show that the [Ne ii] line is optically thin and gas alone cannot be responsible for the blueshift and line asymmetry. Hollenbach & McKee (1989) calculated that a column of Ne⁺ of a few times 10¹⁸ cm⁻² (or a few 10²² H atoms for a solar Ne/H abundance ratio) is necessary to produce an optical depth of 1 at the center of the [Ne ii] line. This density has to be compared with <10¹⁶ cm⁻² Ne⁺ in the EUV layer and <10¹⁷ cm⁻² in the X-ray layer (Hollenbach & Gorti 2009; Glassgold et al. 2007), clearly not sufficient to make the line optically thick. That the line is optically thin is also clear from the observed line profile which does not show any absorption at the line center, where the optical depth is the highest.

Sub-micron dust grains are the next major source of opacity. According to the disk model of Calvet et al. (2002) the inner disk of TW Hya could have only a tiny amount (0.5 lunar masses) of micron-sized grains out to ∼4 AU and thus would be optically thin at visible and even more at infrared wavelengths out to that radial distance. However, Ratzka et al. (2007) show that this model gives mid-infrared visibilities that are a factor of two too high as well as the wrong wavelength dependence of the visibility. They argue that the dust inner rim, hence the transition from thin to thick disk, must happen much closer at about 0.5–0.8 AU. More recently, Akeson et al. (2011) also argue for the presence of optically thick dust within 4 AU. They propose a more complex structure with an optically thick dust ring centered at 0.5 AU, a dust gap, and a full optically thick dust disk starting at ∼4 AU. Our simple approach of combining [Ne ii] profiles from optically thin and thick disk radii suggests that more than 80% of the [Ne ii] emission should arise from disk radii where the midplane is optically thick to the [Ne ii] emission. If the optically thin disk extends out to 4 AU as in the model of Calvet et al. (2002), then the observed [Ne ii] profile cannot be reproduced by the EUV-only and X-EUV models from Alexander (2008) and Ercolano & Owen (2010) because in these models ∼50% of the [Ne ii] emission arises within 4 AU. If the dust inner rim is at ∼1 AU, as in the Ratzka et al. (2007) model, then these models can explain the blueshift and asymmetry in the line profile since they both predict that ∼90% of the [Ne ii] emission comes from outside 1 AU, i.e., from disk radii where the midplane would be optically thick. A different model, also consistent with our observations, is that from Gorti et al. (2011). From a detailed analysis of many gas emission lines from TW Hya and using the Calvet et al. dust distribution with a static disk model, they find that 75% of the [Ne ii] emission line comes from the surface of the optically thick dust disk beyond the 4 AU gap.

Unlike in the case of the [Ne ii] line, the forbidden and semi-permitted optical lines investigated here are not blueshifted with respect to the stellar radial velocity. The lack of blueshift could have two origins: (1) the optical lines trace a wind component at disk radii where the midplane is optically thin to the redshifted outflowing gas or (2) the optical lines do not trace the photoevaporative flow but rather gas that is bound to the star/disk system. In reference to the [O i] optical lines, this means that we do not have an unambiguous diagnostic for the neutral wind component of the disk as we have for the ionized component. In the following, we discuss what other constraints we can place on the disk region traced by the optical lines and what observations could discriminate between the two scenarios presented above.

First of all, the similarity in the FWHM of the [O i] 5577, 6300, 6364 Å, the [S ii] at 4068 Å, and the Mg i] at 4571 Å suggests that these lines trace similar disk radii. Based on the centrally peaked profiles and maximum observed velocities, we argue in Section 4.2 that the emission could extend as close as 0.06 AU to the star, where the disk may be magnetospherically truncated to and possibly beyond 1 AU. The observed [O i] 6300/5577 Å line ratio of ∼7 as well as the non-detection of the [S ii] doublet 6716/6730 Å point to a region of high temperature and electron density. For gas in local thermal equilibrium (LTE) and optically thin emission we can write the [S ii] 6716 and 4069 Å ratio as

$$\begin{equation} \frac{L_{4069}}{L_{6716}} = \frac{n(^{2}P) A_{4069} \Delta E_{4069}}{n(^{2}D) A_{6716} \Delta E_{6716}}, \end{equation} \tag{ 1 }$$

where n(²P) and n(²D) are the upper level populations which in LTE are linked as n(²P)/n(²D) ∼ 4/6 × e^{−14, 000/T}. To explain a line flux ratio higher than six we thus require gas hotter than ∼3000 K. A similar argument applied to the [O i] 6300 and 5577 Å lines yields a similarly high temperature (Gorti et al. 2011). Such a high temperature can be reached within a few AU from the star in a disk atmosphere irradiated by stellar X-ray but LTE requires very high electron densities (n_e ≳ 10⁸ cm⁻³) which may not be present in the photoevaporative wind (Gorti et al. 2011). We turn now to the strong [O i] lines and the neutral component of the disk wind to investigate the two scenarios proposed at the beginning of this section.

In the first scenario, the [O i] emission is thermal and arises from the hot (>5000 K) disk surface heated by stellar X-rays. Ercolano & Owen (2010) have shown that line profiles can be symmetric even if the gas is flowing and any blueshift in the peak emission can be small, a few tenth of km s⁻¹, if most of the emission arises within the dust depleted inner disk.¹⁶ For the specific case of TW Hya an inner dust hole of 2 AU would produce a blueshift in the [O i] line greater than 1 km s⁻¹ and slightly asymmetric profiles while a 4 AU hole would give just a shift of 0.4 km s⁻¹ and a symmetric profile pointing to the flow traced by the [O i] emission extending well beyond 2 AU. The FWHM of the line is in both models relatively large ∼20 km s⁻¹, a factor of two larger than measured. The line broadening can be reduced if the emission in the flow is suppressed and the bound [O i] thermal component arising from within a few AU dominates the emission. This strongly points to the second scenario in which the [O i] lines are not tracing the neutral component of the photoevaporative wind in the disk of TW Hya but rather the bound disk component well inside the gravitational radius.

A variant to the second scenario is that proposed by Gorti et al. (2011). In this model, which is specifically for TW Hya, most of the [O i] emission is not thermal but produced from dissociation of OH molecules by stellar FUV photons leading to O atoms in electronically excited states and their subsequent decay to the ground state. Gorti et al. (2011) find that this mechanism gives a 6300/5577 Å line ratio of ∼7, very similar to the observed one. Because the [O i] emission comes from cooler gas closer to the disk midplane the emitting gas would not participate to the photoevaporative flow and hence [O i] lines would be narrower and symmetric around the stellar velocity. The [S ii] line ratios are also reproduced in this model since most of the emission arises in the hot and dense gas close to the star within the dust depleted inner disk.

A way to directly discriminate between bound and unbound oxygen gas is through spectroastrometry in the [O i] 6300 Å line. For gas in Keplerian rotation the spectroastrometric signal will be symmetric in velocity and the strongest for a slit oriented along the disk major axis, the signal will be zero for a slit perpendicular to the major axis (see, e.g., Figure 12). The presence of a photoevaporative wind alters the symmetric spectroastrometric signal and produces a shift in the flux distribution versus wavelength even for a slit perpendicular to the disk major axis. With a combination of good seeing (∼1'') and high signal-to-noise a few milliarcsecond accuracy can be attained via spectroastrometry at visible wavelengths (see, e.g., Takami et al. 2003). Hence, detecting asymmetries around 1 AU, which would produce shifts of ∼20 mas, is well within the reach of this technique.

Our study of TW Hya represents the first attempt to identify and combine information from multiple gas lines at infrared and optical wavelengths that could trace the photoevaporative disk wind. Here, we summarize the observables reproduced by different disk models as applied to TW Hya and discuss how to nail down photoevaporation rates from protoplanetary disks.

The intensity of the [Ne ii] line at 12.81 μm can be reproduced by all models: in the EUV-only model ionized neon atoms trace the skin of the disk while in the X-EUV and XE-FUV models they trace a deeper, mostly neutral layer that participates to the flow. Any asymmetry in the line profile tells us where most of the emission arises with respect to the optically thick dust disk. A better determination of the transition from thin to thick disk could potentially validate or invalidate the EUV/X-ray photoevaporation models (see Section 5.1). The strong [O i] line at 6300 Å from TW Hya cannot arise from the EUV layer because this layer is mostly ionized and produces only weak [O i] emission. While the [O i] peak emission from several T Tauri stars in the Hartigan et al. (1995) sample presents a small blueshift pointing to a possible origin in the X-ray wind, this is not the case for TW Hya. In addition, the small FWHM of the [O i] line also argues against most of the emission tracing the wind and rather points to bound gas close to the star. The luminosities of the [S ii] at 4068 and the Mg i] at 4571 Å can be reproduced by the X-EUV primordial disk model but with a stellar X-ray luminosity that is 10 times weaker than that of TW Hya (Table 1 from Ercolano & Owen 2010). The non-detection of the [S ii] doublet at 6717 and 6731 Å is also puzzling in the frame of both the EUV and the X-EUV model. For our results to be consistent with these models, sulfur should be depleted in the disk wind. Depletion of more refractory elements such as silicon and aluminum has been previously proposed for the gas surrounding TW Hya (e.g., Lamzin et al. 2004). Elemental abundances therefore add an additional uncertainty when comparing observed line fluxes to model calculations. Alternatively the XE-FUV model developed for TW Hya explains the non-detections of the [S ii] doublet and proposes that the observed [S ii] line at 4069 Å arises from dense, bound gas in the inner disk (Gorti et al. 2011). Similarly to the forbidden optical lines, the CO line profiles and their spectroastrometric signal are consistent with bound gas in Keplerian rotation around TW Hya (Pontoppidan et al. 2008).

More effort should be devoted to identifying gas lines that trace the predominantly neutral winds of XE-FUV and X-EUV irradiated disks both around TW Hya and around other young stars. Establishing the presence of this component would point to flow rates 1–2 orders of magnitude larger than the 10⁻¹⁰  M_☉ yr⁻¹ predicted by EUV-only photoevaporation models. In addition, attempts should be made to empirically measure the ionizing flux impinging on the surface of protoplanetary disks via for instance line flux ratios from two successive stages of ionization of a given element or from volatile elements tracing the flow.