Revisiting the Helium and Hydrogen Accretion Indicators at TWA 27B: Weak Mass Flow at Near-freefall Velocity

TWA 27B (2M1207b) is the first directly imaged planetary-mass (M p ≈ 5 M J) companion and was observed at 0.9–5.3 μm with JWST/NIRSpec. To understand the accretion properties of TWA 27B, we search for continuum-subtracted near-infrared helium and hydrogen emission lines and measure their widths and luminosities. We detect the He i triplet at 4.3σ and all Paschen-series lines covered by NIRSpec (Paα, Paβ, Paγ, Paδ) at 4σ–5σ. The three brightest Brackett-series lines (Brα, Brβ, Brγ) as well as Pfγ and Pfδ are tentative detections at 2σ–3σ. We provide upper limits on the other hydrogen lines, including on Hα through Hubble Space Telescope archival data. Three lines can be reliably deconvolved to reveal an intrinsic width Δv intrsc = (67 ± 9) km s−1, which is 60% of the surface freefall velocity. The line luminosities seem significantly too high to be due to chromospheric activity. Converting line luminosities to an accretion rate yields Ṁ≈5×10−9MJyr−1 when using scaling relationships for planetary masses, and Ṁ≈0.1×10−9MJyr−1 with extrapolated stellar scalings. Several of these lines represent the first detections at an accretor of such low mass. The weak accretion rate implies that formation is likely over. This analysis shows that JWST can be used to measure low line-emitting mass accretion rates onto planetary-mass objects, motivates deeper searches for the mass reservoir feeding TWA 27B, and hints that other young directly imaged objects might—hitherto unbeknownst—also be accreting.


INTRODUCTION
While several hundreds of substellar objects display convincing evidence of ongoing accretion (Betti et al. 2023), only a few have masses below  p ≈ 10  J and have been studied extensively in the last few years: PDS 70 b and c (Wagner et al. 2018;Haffert et al. 2019) and Delorme 1 (AB) b (Eriksson et al. 2020;Betti et al. 2022a,b;Ringqvist et al. 2023).It is therefore capital to enlarge this population to study the dependence of the accretion rate on object mass and age, with the prospect of better understanding the differences in formation processes from stars down to planets.
Brown dwarfs and gas giants likely gain mass through different but not necessarily mutually exclusive physical mechanisms: magnetospheric accretion from a local gas reservoir (e.g., Calvet & Gullbring 1998;Hartmann et al. 2016;Thanathibodee et al. 2019;Hasegawa et al. 2024) or largescale direct accretion onto the surface of the object and its circumplanetary (or "circum-substellar-object disc") disc (CPD) (e.g., Tanigawa et al. 2012;Aoyama et al. 2018;Marleau et al. 2023).These modes of accretion are expected to produce line emission, strongest in the lines of neutral hydrogen.The mass of the accretor is a key determinant of the physical conditions at the accretion shock on its surface or on the CPD.
In turn, these conditions lead to predictions allowing one to distinguish the mechanisms.This is starting to be leveraged (Demars et al. 2023), but line-resolved ( ≳ 15, 000) observations will be required to take advantage of the full diagnostic potential (Marleau et al. 2022).
If only an integrated line luminosity is available, as is often the case, care must be taken to use an appropriate scaling relationship to estimate the accretion rate from the line luminosity (Betti et al. 2023).Blindly extrapolating stellar relationships does not seem justified and in fact likely underestimates accretion rates systematically, by up to a few orders of magnitude (Aoyama et al. 2021;Marleau & Aoyama 2022).Nevertheless, both for spectroscopically resolved and non-resolved (low-resolution or photometric) observations, sensitivity to low line fluxes is critical to avoid being biased towards high fluxes and thus accretion rates; only non-biased observations allow meaningful studies of the scatter in the  ( ) correlation and its dependence on physical parameters such as age (Betti et al. 2023).
In a wider context, the James Webb Space Telescope (JWST; Gardner et al. 2023) has a tremendous potential to help answer outstanding questions in planetary accretion and formation.The advantages of JWST are manifold.Firstly, JWST can access bright hydrogen -transitions, such as Pa  and Br  (respectively at 1.875 μm and 4.051 μm), which are challenging to observe from the ground due to Earth's atmospheric absorption.Traditionally, infrared (IR) observations have played a crucial role as a powerful probe in starforming regions with high extinction, which are not amenable to UV and visible observations.Secondly, medium-resolution spectroscopy ( ∼ 2, 000) using an integral field unit (IFU) with wide wavelength coverage enables the direct detection of multiple IR emission lines in substellar companions.Nearsimultaneous observations are devoid of short-term variability (e.g., GQ Lup B;Demars et al. 2023), enabling the acquisition of reliable results from the analysis of multiple lines.Lastly, JWST provides diffraction-limited observations, resulting in unprecedented sensitivity.This is a particular advantage for observing isolated substellar objects that are generally challenging to observe with ground-based adaptive optics (AO) due to their intrinsic faintness.JWST will characterize more accreting free-floating planets.For reference, the JWST sensitivity at NIR wavelengths is approximately 100 times better than from the ground (see the "Historical Sensitivity Estimates" on the JWST website).These observations are crucial for understanding very faint lines from substellar objects in a quiescent accretion phase (e.g., Brittain et al. 2020) or with very weak accretion close to the end of their formation, which might occur earlier than thought (e.g., Wagner et al. 2023).
With its mass of  p ≈ 5  J , TWA 27B, also known as 2M1207b, is the first planetary-mass directly-imaged object.
It was discovered by Chauvin et al. (2004), two years after identification of the primary as a probably accreting member of the TW Hydrae Association (TWA) by Gizis (2002), and followed four years later by the iconic HR 8799 system (Marois et al. 2008(Marois et al. , 2010)).However, TWA 27B has a mass ratio of only  ≈ 0.2 relative to the primary TWA 27A, which suggests the system might have formed not in a planetary but rather a stellar way (Lodato et al. 2005;Mohanty et al. 2013;Reggiani et al. 2016;Bowler et al. 2020).Luhman et al. (2023) analysed the GTO 1270 (PI: S. Birkmann) JWST/NIRSpec (Jakobsen et al. 2022) integral-field unit (Böker et al. 2022) data on TWA 27B.Thanks to clear hydrogen-line emission, they showed that this object is accreting.They focused on the photospheric emission and measured the flux contained in one hydrogen transition, Pa .Here, we search systematically for hydrogen lines and report the continuum-corrected line fluxes.We then use scaling relationships between line and accretion luminosity to estimate the accretion rate, and argue robustly that the line fluxes are not dominated by chromospheric activity.

Data sources
The JWST data were presented in Luhman et al. (2023) and were obtained with the NIRSpec instrument.They consist of one-hour exposures starting on 2023 February 07 at 17:52:28.779with the G395H/F290LP highresolution-grating-filter combination (0.97-1.89 μm; after this overview the filter names will be dropped for conciseness), at 18:32:50.598with G235H/F170LP (1.66-3.17μm), and at 19:11:59.463with G140H/F100LP (2.87-5.27μm).The spectrum for each half-grating and the errorbars were kindly shared by K. Luhman.Some of the flux errorbars were spuriously high and we replaced them with the median of the well-behaved errors.In any case, the dominant source of noise is the continuum noise, as shown below.
In addition to JWST data, we provide an H  photometric point from our analysis of Hubble Space Telescope (HST) archival data.HST observed the TWA 27 system with WFC3/UVIS on 2011 March 28 (PID: 12225, PI: A. Reiners) and we show the F656N image in Appendix A. The HST image does not yield a significant detection of TWA 27B (only a tentative one) and provides an upper limit on its H  emission.We will use this in Section 3.2.

Fitting approach
We searched the JWST data for a signal at the metastable neutral helium triplet He i 10833 as well as all hydrogen lines covered by the detectors in the Paschen-transition series (electron final energy level   = 3), which is   = 4-7 (Pa -Pa ); all lines up to an initial level   = 10 (Br 10) in the Brackett series (  = 4); and all lines up to   = 10 (Pf ) in the Pfund series (  = 5), but without Pf  (  = 6) since it is off the red edge of NIRSpec with  0 = 7.46 μm.An overview of the lines is in Table 1.From the Humphreys series (  = 6), only the transitions   = 10-12 or 13 (5.128-4.170or 4.376 μm, respectively;   = 13 is on the detector edge) are covered by NIRSpec and visual inspection did not reveal any significant peak at the respective locations.Therefore, we do not analyse this series.
In Figure 1a, we show as an example the Pa  line, covered by G140H and G235H.We show the data from each grating but also the data from G140H downgraded to the resolution of G235H.For this, we convolved the G140H data with a Gaussian of width Δ given by Δ 2 = Δ 2 2 − Δ 2 1 , where Δ 1 (Δ 2 ) corresponds to the resolution of G140H (G235H) at Pa ,  = 3769 ( = 2129), as reported in Table 1.As a reminder, the Gaussian function used for smoothing and as the line profile shape in the analysis is  =  0 exp [−0.5(/Δ) 2 ], where  is the filter height or the flux density and  is the Doppler velocity offset from the vacuum central wavelength.When compared at similar resolution, the flux levels are mostly consistent in absolute value.A few features are seen in the downresolved G140H data but not in G235H, suggesting that the errorbars are still somewhat estimated.However, the overall agreement is very good.
For each line, we fitted the local continuum by a cubic function and the line excess by a Gaussian profile, using gnuplot's built-in fit routine.We took a three-step approach: 1. We fitted the continuum in a range Δ cont = 2700 km s −1 on either side of the theoretical vacuum line centre (Wiese & Fuhr 2009), masking out || ≡ Δ mask = 300 km s −1 .This value of Δ cont prevents the He i and Pa  lines, separated by only about 2900 km s −1 , from interfering with each other, and Δ mask is a few times the expected line width for an  p ≈ 5  J ,  p ≈ (1-2) J object (Luhman et al. 2023), which is conservatively at most Δ ≈  ff, ∞ (Aoyama et al. 2018), where  ff, ∞ = √︁ 2  p / p ≈ 120 km s −1 is the free-fall velocity from infinity.For two lines, the grating edge shortened slightly the actual wavelength range used for the continuum on the red side (Br  at +1300 km s −1 and Pa  at +2200 km s −1 ), compared to the nominal Δ cont = 2700 km s −1 , but this is inconsequential.
An important outcome of this fitting is the root-mean square (RMS) deviation from the fitted continuum (more properly, the standard deviation, but we adopt the common usage).This continuum RMS is a source of noise because we do not attempt to identify and remove spectral features.
2. We fitted each continuum-subtracted line with a Gaussian function, again taking only the bin errorbars into account.
3. We initialised the fit parameters of the continuum and line excess to the values from the sequential fits and did a joint fit.For the errorbars of each bin, we added in quadrature to the errorbar of each bin the RMS of the continuum.The fits converged in around ≈ 5-20 iterations, typically around 10.
An example outcome is shown in Figure 1b.We varied Δ cont and Δ mask and obtained essentially the same fit results for all the lines (not shown).We note that the continuum RMS dominates the noise budget (see the pale and dark grey errorbars in Figure 1b).
As a comparison, we also performed the fits using only the flux uncertainty in each bin as the errorbar.Some fitted lines were narrower than the instrumental broadening, but within only 1, and the overall results were very similar.Other variations in the approach also led, if at all, to small differences of at most a few km s −1 in the line widths.Thus the precise treatment of the errorbars does not matter in this case.
This fitting approach yields line fluxes, always meant as an excess above the fitted continuum (see Section 3.1).The integrated luminosity of a given line is  line = 4 2  line , with  = 65.4 pc the distance to TWA 27B (see Luhman et al. 2023 and references therein) and  line = √ 2 × Δ ×  0 / × F 0 , where Δ is the fitted Doppler-shift velocity width, F 0 is the flux-density peak value in excess of the continuum,  is the speed of light, and  0 is the central vacuum wavelength of the transition.
For simplicity and clarity, we do not attempt to correct the fluxes for possible extinction.Interstellar extinction towards the TW Hydrae Association (TWA) and the TWA 27 system is negligible (Herczeg et al. 2004;Mohanty et al. 2007).
Extinction by an edge-on disc around TWA 27B has been suggested to reconcile the tension between the SED-derived  eff and theoretical predictions (e.g., Mohanty et al. 2007).However, the analysis of Luhman et al. (2023) suggests that using other atmospheric models instead might obviate the need for heavy extinction.Otherwise, somewhat "tuned" cloud properties could explain the spectrum and brightness of TWA 27B (Skemer et al. 2011).The ultimate answer is not settled, and the line-emitting regions might be differently extincted than the atmosphere.Thus keeping the fluxes "as is" therefore avoids introducing uncertainty.

Detections and non-detections
In this work, we qualify a line as detected only if it has more than two bins at more than 3.This is less stringent than the often-used 5 criterion but appears justified given Figure 1.Left: The Pa  line through the red half of the G140H (red) and the blue half of the G235H (blue) gratings at full resolution (pale red and bright blue), and from G140H down-convolved and re-interpolated to the resolution of G235H (bright red).Right: He i data (black) with errorbars (dark grey: only from each bin; pale grey: adding the continuum RMS in quadrature), cubic fit to the continuum (grey dashed line), and fit to the continuum and line (purple).The continuum RMS noise is shown in Figure 7.
the low level of surprise, the modest impact of a detection, and the negligibility of the "Look-Elsewhere Effect", to use the considerations discussed by Lyons (2013).Unaccounted-for systematics will reduce the true significance of the detections but only modestly since the systematics seem small.
Other lines detected at 2-3 will be considered only tentative detections.We report their fit parameters but do not further analyse them.Finally, for the other lines for which there is clearly no signal, we compile only upper limits on the flux and luminosity, and no fit parameters.We calculate upper limits by using an upper-tail one-sided test based on the 2 distribution.Specifically, we solve for the minimum line flux which, broadened to the resolution at that wavelength, would lead to a 3 deviation from a null excess: where Δ inst = //(2 √ 2 ln 2) is the effective line width given by the instrumental resolution1 ( 0 ),  F cont is the RMS of the fitted continuum (i.e., its standard deviation), and  2 crit ≈ 33 is the critical value of the  2 distribution that corresponds to 3 (probability of 99.7 %) for  ≈ 14 − 1 degrees of freedom, since around 13-14 spectral bins are found within a (somewhat arbitrarily-chosen) range of ±3Δ inst around  0 .We compute  for each line (note the squared exponential) and always find  ≈ 1.6-1.7.Thus  upp line ≈ 11.4 ×  F cont × Δ inst  0 /.A more detailed analysis could use the injection of fake planets and take the 1 Obtained from the (pre-launch; dated 2016 August 30) "Dispersion curves for the NIRSpec dispersers" section of the JWST website.
spatially varying sensitivity into account (e.g., Bonse et al. 2023).However, since we will not analyse the non-detections in detail, we keep the simpler approach here.It is in principle more accurate than but comparable to the expression of Betti et al. (2022a), depending on the transition) is their spectral dispersion and  pix = (7 Å)/Δ disp , or to the expression of Alcalá et al. (2014) or Gangi et al. (2022),  upp line = 3Δ ×  F cont with Δ = 1-2 Å their assumed line width.

Line fluxes and widths
We detect the neutral helium triplet at 4.3 and all Paschenseries lines covered by the observations (Pa , Pa , Pa , Pa ) at > 3.5-5.The three brightest Brackett-series lines (Br , Br , Br ) as well as Pf  and Pf  are tentative detections at 2-3.For the remaining hydrogen lines accessible to NIRSpec, we obtain upper limits.All detections and nondetections are presented in Table 1, and illustrated in Figure 2, with line profiles in Appendix B. One of the lines with only an upper limit is Pf  (on both gratings that include it), which is formally detected at 1-2.1 but whose spectral appearance is clearly not credible (see last figure of Appendix B: Figure 17).All multiply-detected upper limits are very similar between both detectors.
The only quantitative point of comparison with Luhman et al. (2023) concerning the line analysis2 is the integrated Pa  line flux.They report  ★ line = (6.6 ± 1.2) × He Note-We consider all hydrogen lines covered by NIRSpec, but limited to an initial level   ⩽ 10 in the Brackett (  = 4) and Pfund (  = 5) series.The subscripts indicate the grating (1: G140H, 2: G235H, 3: G395H), with Pa , Br , Br 10, and Pf  covered by two gratings.F cont is the fitted continuum level at  0 , with uncertainty  F cont given by the RMS computed over Δ cont = ±2700 km s −1 (see text). line and  line are the continuum-subtracted, line-integrated flux and luminosity. is the spectral resolution at  0 .   is the significance of the integrated line flux or the peak significance; the peak flux density is Δ and  0 are the fitted Doppler line width and velocity offset of the peak.R Δ is the fitted line width relative to the instrumental broadening; values below one suggest the continuum is overestimated.For lines clearly not detected (see Section 2.3),  line and  line are 3 upper limits (see text).
10 −17 erg s −1 cm −2 , which however includes the contribution of the continuum, which we indicate with the ★ superscript.Our errorbars are similar but only half as large.Our continuum-subtracted flux is smaller:  Pa  = (2.7 ± 0.6) × 10 −17 erg s −1 cm −2 .Taking the continuum to contribute simply Δ = F cont × (Δ) 0 /, where Δ is the fitted width, we obtain √ 2 before adding it to the line excess.Thus we agree in the recovered flux within the errorbars.
We note the following points: • The Pa  line is detected with G140H ( ≈ 3700) and G235H ( ≈ 2100), as shown in Figure 1a.The integrated fluxes are almost equal, differing only by the quadratic sum of the 1 errorbars.The G235H line is nominally narrower by roughly 2 but this reflects the higher fitted continuum level, even though it agrees within less than 1 with the continuum fitted on G140H.The true continuum level is unknown, and Figure 1a reveals that some features are absent at lower resolution, but it is unknown whether these features are real or only noise.
• At the other lines covered by two detectors (Br , Br 10, Pf ), the respective continuum level and its uncertainty are the same to much less than 1 between the two detectors.This suggests that the continuum is not heavily variable on a 40-min timescale and that there are no strong systematics between the detectors.
• For Pf , a strong dip near +2200 km s −1 could lead to an overestimate of the RMS and could thus reduce the significance of the line.
In Figure 3a, we plot the line widths for the clearly-detected lines (He i and the Paschen series).Pa  from G235H is  1, coloured by series.Thick errorbars are for clear or tentative detections, and thin arrows for 3 upper limits (see text).Pa  2 and Br  1 are shifted for clarity.The 3 upper limit on H ,  line < 6.4 × 10 −9  ⊙ , is not shown.The spectral density F (at full resolution and smoothed; pale and dark grey, respectively) is shown against an arbitrary linear flux scale.The small gaps in the spectral coverage are detailed in Luhman et al. (2023).
slightly narrower than the instrumental resolution, but by less than 1, which suggests that the continuum level is slightly over-and the line strength underestimated.However, all other lines (including Pa  from G140H) are wider than the instrumental resolution, which means that they are somewhat resolved.In fact, assuming that both the intrinsic line shape and the instrumental PSF are Gaussians, we can easily estimate the intrinsic line width (that is, deconvolve) by quadratic subtraction: Δ 2 intrsc = (Δ) 2 − Δ 2 inst .This gives Δ intrsc = (57 ± 20) km s −1 for He i, 76 ± 14 for Pa , 28 ± 22 for Pa , 41 ± 32 for Pa , 62 ± 15 for Pa  1 (dropping the units), where the errorbars come from error propagation assuming negligible uncertainty on Δ inst .Since the fitted line width is less than 1 above the instrumental resolution at Pa  and Pa , we will not consider them for the analysis now.The inverse-variance weighted average of the deconvolved widths is ⟨Δ intrsc ⟩ = (67 ± 9) km s −1 .We will discuss this in Section 4. For reference, the inverse-variance weighted average of the fitted line widths is ⟨Δ⟩ = (73 ± 5) km s −1 .The mean deconvolved line width is only 1.5 smaller than this.However, ⟨Δ⟩ is in principle not a meaningful quantity because each line is instrumentally broadened by a different amount.Therefore, we will consider only Δ intrsc .
The velocity zero-points  0 (that is, the Doppler offsets from the vacuum central wavelength) are shown in Figure 3b.The shorter-wavelength lines (He i, Pa , Pa , Pa ) are less con-sistent with zero, with an average near  0 ≈ +(20±10) km s −1 .This, and the  0 of the Pa  line from either grating, is close to or consistent with the radial velocity of the primary of  = +(11 ± 2) km s −1 (Faherty et al. 2016), which is likely the systemic velocity.

Accretion luminosity
In Figure 4, we compare the accretion luminosity  acc derived from each line independently.We use the H  upper limit (see Appendix A),  < 1.6 × 10 −17 erg s −1 cm −2 , which we multiply by three to estimate the 3 limit.We use  acc ( line ) relationships of the form log 10  acc / ⊙ =  log 10  line / ⊙ + , where (, ) are fit coefficients.As in Betti et al. (2022a,b), we use both the relationships for CTTSs extrapolated down to the line luminosities measured at TWA 27B, and also the  acc ( line ) relationships based on detailed shock models designed for planetary-mass objects (Aoyama et al. 2021;Marleau & Aoyama 2022).We will refer to the latter as "planetary  acc ".Magnetospheric accretion columns could contribute to the line fluxes also for planetary-mass objects but these models do not include this.We note that the Aoyama et al. (2018) model does not make predictions for helium lines.
Specifically, for the CTTS scalings we use the coefficients in Alcalá et al. (2017) for all lines available, complementing with Salyk et al. (2013) for Pa , Alcalá et al. (2014) for He i, and Komarova & Fischer (2020) for Br .Most coefficients are summarised in Aoyama et al. (2021) and compared to the planetary values.We note that no CTTS scalings exist for Pa , Br , Br , and Pf .
We propagate the error from the fit and the uncertainty on the line flux.For CTTSs, the fit error is from the errors on  and , and for the planetary relationships we fix  = 0.3 dex (see Aoyama et al. 2021), added through error propagation to  from the line flux.
The inverse-variance-weighted average of the planetary  acc from the detections at Pa -Pa  is ⟨log( acc / ⊙ )⟩ = −5.23 ± 0.14 dex (blue dashed line and 1 band in Figure 4).Even though the errorbars are small ( ≈ 0.2 dex), the  acc values from the individual lines lie within 1 of the average.With the exception of H , to which we will return below, the upper limits from the other lines are consistent with this.When using the extrapolated CTTS scalings, the errorbars on  acc are larger ( ≈ 0.6 dex), but there too, the inferred  acc from each line agrees within 1 with the average ⟨log( acc / ⊙ )⟩ ≈ −7 dex (not shown).The major difference is that this is roughly 1.7 dex smaller than the average  acc as inferred from planetary models.This is in line with the conclusions of Aoyama et al. (2021) or, for Delorme 1 (AB) b, Betti et al. (2022a,b).
Interestingly, the  acc corresponding to the 3 upper limit on  H  is lower than the average  acc by 1 or 0.5 dex according to the planetary or CTTS relationships, respectively.This apparent discrepancy could be explained by time variability (the H  was measured in 2011, the other lines in 2023), wavelength-dependent absorption due to dust (see Figure 9b of Marleau et al. 2022, but given the low accretion rate, absorption is likely negligible), or differences in the emission mechanism for H  compared to the other lines (at least for TW Hydra, H  does not correlate as well as the other lines with  acc ; Herczeg et al. 2023).No strong arguments can be made for or against these factors.Otherwise, instrumental or other observational effects (including systematics) need to be invoked, but we assume that they cannot explain most of the 0.5-1 dex difference.
In Figure 4, we also applied the extrapolated He i  acc ( line ) scaling relationship for CTTSs, which was studied only by Alcalá et al. (2014).The helium triplet is very sensitive to both winds and accretion (e.g., Fischer et al. 2008;Thanathibodee et al. 2022;Erkal et al. 2022) and should be used with caution, as Alcalá et al. (2014) note.The modest resolution of the NIRSpec data,  = 2062, does not allow for detailed studies of the line shape (see the types proposed in Thanathibodee et al. 2022), especially since the line width is barely wider than the instrumental broadening (Figure 3a; profile in Fig. 7).Nevertheless, the  acc inferred from He i is consistent with the average from the other lines.High-resolution line profiles could reveal whether this is a coincidence, and theoretical predictions extending the Aoyama et al. (2018) models would be welcome.

Accretion rate
Combined with an inferred mass of  p ≈ 5  J and  p = 1.4  J from Luhman et al. (2023), the average  acc can be translated into an "accretion rate" , whose meaning requires discussion (see afterwards).From the planetary scalings, we obtain  ≈  acc /(  p / p ) ≈ 5 × 10 −9  J yr −1 .Using instead the extrapolation of the Alcalá et al. ( 2017) CTTS relationships would imply  ≈ 0.1 × 10 −9  J yr −1 , while the scaling of Natta et al. (2004) applied to our  Pa  value yields  ≈ 0.3 × 10 −10  J yr −1 .This is slightly lower than the  ∼ 10 −10 -10 −9  J yr −1 that Luhman et al. ( 2023) report using that relationship.The reason is that these scalings are defined for continuum-subtracted (i.e., photosphericemission-corrected) values, while Luhman et al. (2023) took the total line flux, leading to an overestimate by around 0.4 dex.This is comparable to the scatter in these relationships and clearly smaller than their systematic uncertainties (Betti et al. 2023), and thus not a major issue.The targets analysed in Alcalá et al. (2017) have Pa  luminosities  line ∼ 10 −6.5 -10 −3  ⊙ .Thus, applying this relationship to TWA 27B requires an extrapolation by 2 dex.This is only slightly smaller than the range of data and therefore possibly acceptable, but it would require some validation.
This accretion rate, whether using the planetary scalings or the extrapolated CTTS ones, is only an estimate due to uncertainties in the prefactors (for instance due to a finite starting radius for the infall).However, even the presumably more realistic, and much higher, value from the planetary scalings implies an "accretion rate" that is small in the sense that the nominal mass doubling time  =  p /  ∼ 1 Gyr, which is orders of magnitude longer than the age of the system (10 ± 2 Myr; Luhman 2023) or of the TWA.Thus TWA 27B would be at the very end of its formation.This is in line with the finding that  acc is smaller than the bolometric luminosity (see Figure 4) by about 0.4 dex according to planetary scalings, and even more if extrapolating the CTTS scalings.While of interest for observers and theoreticians alike, the accretion processes seem to be currently of very subdominant existential relevance for TWA 27B.
However, as discussed in Marleau et al. (2023), this  is not necessarily the growth rate of the planet but rather only what is hitting the planetary surface and/or the CPD close to the planet: in practice, gas needs to shock at  ≳ 30 km s −1 to emit lines3 (Aoyama et al. 2018).TWA 27B might also be accreting through a boundary layer, which likely does not generate lines.Then, the total accretion rate would be larger than inferred.Given the mass ratio with TWA 27A ( ≈ 0.2) and their separation (55 au), it is unclear whether physical scales comparable to or larger than the Hill sphere  Hill ≈ 22 au are feeding TWA 27B, or whether it can draw the matter it is accreting only from a CPD.There is no clear published IR flux excess (while, as Luhman et al. 2023 note, there might be in unpublished 5-28 μm MIRI data), and the Atacama Large Millimeter/submillimeter Array (ALMA) upper limit from Ricci et al. (2017) constrains little the amount of mass available in the system or the size of the mass reservoir.On the other hand, if TWA 27B is undergoing magnetospheric accretion and the columns contribute to the line emission (Hartmann et al. 2016), the line flux may indeed be tracing (nearly) the whole mass flow.
We keep these caveats in mind but compare in Figure 5 this accretion rate with those of the other TWA members.The  values are from Venuti et al. ( 2019), with TWA 30B excluded because it is likely severely affected by extinction (Looper et al. 2010;Rodriguez et al. 2015).Especially towards higher-mass objects, the line luminosities might trace the entire mass flow, whereas for the lower mass objects (TWA 27A and TWA 28, since no lines were detected at TWA 40), the reported  values might be lower limits.
Intriguingly, the accretion rate onto TWA 27B according to the planetary scaling relationship is somewhat smaller than the CTTS-scalings-based  for TWA 27A, with an instantaneous ratio  inst ≡  B /(  A +  B ) ≈ 0.2, where  A ≈ 10 −7.8  J yr −1 (Venuti et al. 2019).At a mass ratio  ≈ 0.2, different hydrodynamics simulations predict on average a steady-state  ≈ 0.6-0.9(Bate 2000;Lai & Muñoz 2023).That the theoretical and observational  are only within a factor of a few from each other suggests that the  estimate using the planetary paradigm for TWA 27B might be relatively accurate.However, the measurements are not contemporaneous (TWA 27A: taken in 2010 and 2012; TWA 27B: 2023), while accretion-rate variability is expected in binary systems on several timescales (e.g., Muñoz et al. 2020).This comparison should therefore be repeated after further monitoring.

DISCUSSION
We briefly discuss some aspects of our results.Line width.-Wefound an average deconvolved (intrinsic) line width Δ intrsc = (67 ± 9) km s −1 based on the wellresolved lines He i, Pa , and Pa .The other clearly-detected lines, Pa  and Pa , have a fitted width barely wider than the instrumental resolution.Shock models predict the intrinsic line width to be not directly equal to but of the order of the preshock velocity (Aoyama et al. 2018;Aoyama & Ikoma 2019).The exact line width depends on the preshock conditions because they set where (at what depth and thus temperature) in the postshock region a particular line is formed.Hydrogen and helium lines might originate from different depths given the different excitation energies of the starting levels.Nevertheless, all intrinsic line widths and the average are consistent with each other to within 1.
If  p = 5  J and  p = 1.4  J (Luhman et al. 2023), the surface freefall velocity is  ff, ∞ = 117 km s −1 , so that Δ intrsc ≈ 0.6 ff, ∞ .Qualitatively, this therefore seems to be in line with shock-model predictions.This also agrees with simulations of accretion onto gap-opening planets, which show that while most gas falling from Hill-sphere scales lands on the CPD and not on the planetary surface (Tanigawa et al. 2012), the contribution from the planetary surface should dominate (Marleau et al. 2023).At the same time, this line width also seems broadly consistent with predictions from magnetospheric accretion (Thanathibodee et al. 2019).Thus, only a medium-resolution line profile is not sufficient to distinguish the two scenarios, and both higher-resolution observations and quantitative modelling are required (Demars et al. 2023).However, this Δ intrsc matches well the interpretation that TWA 27B is not a higher-mass (substellar) object with an inclined disc leading to high extinction (see Luhman et al. 2007) since the higher mass would likely lead to a wider line than observed.
Chromospheric activity?-Accretion at low-mass objects is a barely-charted territory and a valid concern is whether the observed line emission comes from accretion or chromospheric activity.Figure 5 shows that the  of TWA 27B is 1.5 dex above the chromospheric noise limit (Manara et al. 2013(Manara et al. , 2017;;Venuti et al. 2019) at  p = 20  J .How the limit behaves at lower masses, that is, closer to TWA 27B's  p ≈ 5  J , is an open question; it might extrapolate as a powerlaw or drop precipitously below a certain mass.However, a sharp rise by several orders of magnitude seems unlikely.Therefore, despite the statistical uncertainties in , accretion appears to be the likeliest source of the lines at TWA 27B.A similar argument can be made for Delorme 1 (AB) b (Eriksson et al. 2020;Betti et al. 2022a,b;Ringqvist et al. 2023).
Helium line emission-With its clear He i 10833 emission (Luhman et al. 2023), TWA 27B joins the select club of known accretors below ≈ 20  J exhibiting He i lines.

SUMMARY AND CONCLUSION
We have re-analysed the processed JWST/NIRSpec spectrum of TWA 27B presented in Luhman et al. (2023), focussing on the accretion indicators because they stated detections of three or four Paschen lines and of the He i triplet.We searched systematically for all accessible hydrogen lines, fitted and subtracted the continuum, and measured line shapes and total fluxes.We quantified the uncertainty on the line shape and integrated flux by measuring the "photospheric noise", which is the continuum residual from the (not-modelled) atmosphere.Our results are: 1.He i is detected at > 4 and the Paschen-series lines that NIRSpec covers (Pa -Pa ) at (3.5-5).These are robust detections.The Br , Br , and Br  lines are tentatively detected with (2-3).The Pf  and Pf  signals are marginal.For the other lines (Br -Br 10, Pf , and Pf ), we obtain upper limits.
2. The Pa  line is covered by two grating-and-filter combinations, which observed TWA 27B ∼ 40 min apart.The similarity of the continuum and the line shape between both observations suggests little continuum variability and no strong systematics between the gratings.
3. We independently fit each line and find that for the lines clearly detected at > 3, namely He i, Pa  1 , and Pa , the fitted widths are well above the instrumental resolution (especially the latter two thanks to  ≈ 3800 and ≈ 2500, respectively).Their de-convolved (intrinsic) widths are consistent with each other and averaging to Δ intrsc = (67 ± 9) km s −1 .This is around 60 % of the free-fall velocity at the surface of TWA 27B.
The width is qualitatively consistent with shock-model predictions (Aoyama et al. 2018) and the result that the planetary-surface shock and not the CPD-surface shock should dominate the emission (Marleau et al. 2023), but also with expectations from magnetospheric accretion (Thanathibodee et al. 2019).Detailed modelling is required to relate preshock velocities and line widths and to help determine where and how the emission lines are formed.
5. Using scalings derived for CTTS (e.g., Alcalá et al. 2017) extrapolated to planetary luminosities or scalings based on detailed models yields accretion luminosities that are consistent between the different lines but discrepant by 1.7 dex between the two approaches, as found before (Betti et al. 2022a,b).Correspondingly, the line-producing (shock) gas mass flow rate  ≈ 5 × 10 −9  J yr −1 is about 50 times higher according to planetary scalings than with the extrapolated CTTS relationships.If  is the growth rate of TWA 27B, its formation is over.The possibility of magnetospheric accretion and other considerations introduce uncertainties about the meaning of .
6.All integrated line luminosities are on the order of  line ∼ 10 −9  ⊙ .Despite theoretical uncertainties, the lines detected come robustly from accretion processes and not from chromospheric activity.
A detailed study of the atmospheric properties could help identify atomic and molecular features in the spectrum of TWA 27B.Removing them would reduce the continuum noise, improve the accuracy of the inferred line shape and flux, and increase the statistical significance by decreasing the noise.Thus the significance of our detections might be currently underestimated.Deeper and spatially better resolved ALMA observations would be very valuable to constrain the amount of mass available in the TWA 27 system and the location of the gas reservoir from which TWA 27B is drawing.
Twenty years after its discovery, TWA 27B still holds many surprises.It could well be that also other young directlyimaged companions are accreting even though no CPD has been detected so far and even if the parent disc is long gone.A deep look with JWST could be worthwhile, and our results suggest that He i, Pa , and Pa  might be particularly wellsuited tracers.
We are indebted to Kevin Luhman for impressively fast and helpful comments and answers and for generously sharing his data.We

A. HST DATA AT H 𝛼
In Figure 6, we show the primary-subtracted F656N image (that is, at H ) of the TWA 27 system taken by the Hubble Space Telescope with WFC3/UVIS2.TWA 27 was observed on 2011 March 28 (PID: 12225, PI: A. Reiners), with an integration time of 120 s.The location of TWA 27B is indicated by a red box.The flux at the position of TWA 27B, measured in an  = 2.5 pixel aperture, is 2.2 above the background, which was calculated as the standard deviation from  = 2.5-pixel apertures at the same angular separation as TWA 27B.Therefore, the small excess is tantalising but a confident H  detection (> 3) would require additional observations.We use the flux upper limit in Section 3.2.

B. ALL LINE PROFILES
In Figures 7-17 we show profiles of the He i and the hydrogen lines for which we searched.In each case we show a broader region with the fitted continuum (top panel) and the continuum-subtracted profile in a zoomed-in region (bottom).The integrated significance of each line, using    from Table 1, is indicated in red below each top-panel curve.

Figure 2 .
Figure 2. Summary of the line luminosities in Table1, coloured by series.Thick errorbars are for clear or tentative detections, and thin arrows for 3 upper limits (see text).Pa  2 and Br  1 are shifted for clarity.The 3 upper limit on H ,  line < 6.4 × 10 −9  ⊙ , is not shown.The spectral density F (at full resolution and smoothed; pale and dark grey, respectively) is shown against an arbitrary linear flux scale.The small gaps in the spectral coverage are detailed inLuhman et al. (2023).

Figure 3 .
Figure 3. Analysis of the clear detections.Top: Fitted line widths Δ (filled circles) and instrumental resolutions Δ inst (open triangles), and deconvolved (intrinsic) line widths Δ intrsc for He i, Pa , and Pa  1 (filled squares, connected by dashed lines to Δ).They are blueshifted and Pa  2 is red-shifted for clarity.Bottom: Line centroids.The radial velocity of the primary (Faherty et al. 2016) is shown as a dashed green line with shaded error region.

Figure 4 .
Figure 4.The accretion luminosity of TWA 27B based on scaling relationships for CTTSs (open grey stars symbols, extrapolated;Salyk et al. 2013;Alcalá et al. 2017) and for planets (filled blue circles;Aoyama et al. 2021; Marleau & Aoyama 2022).Arrows (same colours; thin and thick, respectively) show the 3 upper limits of Table1, with Br  for CTTSs(Komarova & Fischer 2020) outside the plot because it predicts  acc ∼ 10 −10  ⊙ .The  acc from He i (dark red for clarity) is fromAlcalá et al. (2014), without a planetary-scaling counterpart.We display the weighted mean of the  acc inferred from planetary-shock-model scalings (blue dashed line and band).The bolometric luminosity (gold; fromLuhman et al. 2023) is shown for comparison.

Figure 5 .
Figure 5. Accretion rates of TWA objects.We show the data of Venuti et al. (2019) (grey stars; multiple observations are joined by a line; only TWA 1 is off the plot with  ≈ 600  J ) and add our analysis of TWA 27B using the Aoyama et al. (2021) relationships (filled circle) or the extrapolated CTTS relationships (open circle).Labels give the TWA numbers.The blue band indicates the chromospheric noise limit at 3-10 Myr from Venuti et al. (2019).
thank Tomas Stolker for help with the flux errorbars; Paul Mollière and Gabriele Cugno for comments on JWST; and Sarah Betti for discussions of upper limits.G-DM acknowledges the support of the DFG priority program SPP 1992 "Exploring the Diversity of Extrasolar Planets" (MA 9185/1).G-DM also acknowledges the support from the Swiss National Science Foundation under grant 200021_204847 "Plan-etsInTime".JH is supported by JSPS KAKENHI Grant Numbers 21H00059, 22H01274, and 23K03463.YA is funded by China Postdoctoral Science Foundation (2023M740110).Parts of this work have been carried out within the framework of the NCCR PlanetS supported by the Swiss National Science Foundation.This work is based in part on observations made with the NASA/ESA/CSA James Webb Space Telescope.The data were obtained from the Mikulski Archive for Space Telescopes at the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS 5-03127 for JWST.These observations are associated with program GTO 1270.This research is based in part on observations made with the NASA/ESA Hubble Space Telescope obtained from the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS 5-26555.These observations are associated with program 12225.Some of the data presented in this paper can be obtained from the Mikulski Archive for Space Telescopes (MAST) at the Space Telescope Science Institute.The specific HST and JWST observations analysed can be accessed via DOI: 10.17909/2f9b-ea80.This research has made use of NASA's Astrophysics Data System Bibliographic Services.Figures 1-5 were produced using GDL, an actively-developed open-source drop-in alternative to IDL.

Figure 7 .
Figure7.Line profile for He i.We show the fitted continuum and line (black) and just the continuum (blue line).Top: whole range used for calculating the continuum, excluding the points in red (close to the line centre).Blue band: ±1 range (RMS of continuum).Bottom: zoom-in on the continuum-subtracted line.Bands: (1, 2, 3), dotted grey lines: (4, 5).Pink: instrumental broadening.The dark grey part of the errorbars: only the error on the bin (as we re-determined it); full errorbar: adding the continuum RMS in quadrature.

Figure 8 .
Figure 8.As in Figure 7 but for Pa  on the two gratings: red half of G140H (top), blue half of G235H (bottom).

Figure 9 .
Figure 9.As in Figure 7 but for Pa  and Pa .

Figure 10 .
Figure 10.As in Figure 7 but for Pa , the highest-order potentially detectable Paschen line.

Figure 12 .
Figure 12.As in Figure11but for Br , which is tentatively detected, and for Br , which is a clear non-detection.For non-detections, the height of the instrumental Gaussian is set to 3 for illustration purposes.

Figure 15 .
Figure 15.As in Figure 11 but for Pf  (not detected) and Pf  (tentatively detected, but with more significance if the continuum is in fact lower).

Figure 16 .
Figure 16.As in Figure 15 but for Pf , very tentatively detected.

Figure 17 .
Figure 17.As in Figure15but for Pf  from two gratings (red half of G235H on top, blue half of G395H on the bottom), in both not detected.
Figure 11.As in Figure 7 but for Br  and Br .