He I $λ$10830 Resonant Scattering in M17: Toward a Diagnostic of Optical Depth and Gas Kinematics

Introduction

The He I $\lambda10830$ emission line, one of the strongest in the near-infrared, is observed in H II regions, planetary nebulae, and active galactic nuclei. The lower level of the transition producing this line, i.e., He 0 $1s2$ $^{3}S_{1}$, is a metastable state. Consequently, despite the excitation potential being as high as $\sim20$ eV, helium atoms in such an excited state are expected to be highly abundant in nebular environments, and the He I $\lambda10830$ line can become optically thick, so that its photons are resonantly scattered. Using Monte Carlo radiative transfer calculations for a static, spherically symmetric gas, B. T. Draine [6] predicted a characteristic distortion of the He I $\lambda10830$ profile by resonant scattering. Unlike hydrogen Ly$\alpha$, which is also strongly affected by resonant scattering, this behavior arises because the He I $\lambda10830$ triplet components with $J=0$, $1$, and $2$ are located at very closely spaced wavelengths. As a result, as the optical depth increases, multiple peaks are expected to appear in the He I profile, accompanied by an overall blueshift. This offers the potential for a new method of diagnosing the physical conditions in nebulae from emission-line profiles, but verifying it requires a near-infrared spectrograph with $R \equiv\Delta\lambda/\Delta\lambda\gtrsim10{,}000$. Such resolution had been reached long ago: with a Fabry–Pérot interferometer at $R \approx36{,}000$, A. H. Vaughan [22] found P Cygni-like profiles of the line in

Email: [email protected] planetary nebulae and the Orion Nebula and attributed them to scattering in expanding gas, before such profiles had been calculated. Prompted by D26, Y. I. Izotov et al. [15] found He I λ10830 broader than Hβ at $R = 5600$ in compact star-forming galaxies and attributed the broadening in part to resonant scattering and in part to turbulence in denser inner regions.

M17, also known as the Omega Nebula, is a giant H II region in our Galaxy and is an ideal target for this test. The dominant ionizing source is CEN 1, a pair of O4 V stars [5, 12], whose hard radiation field ionizes helium as well as hydrogen. Combined with the high densities of the southwestern (SW) bar ($n_e \sim10^2$–$10^3$ cm$^{-3}$; [19]), this implies a large He$^0$ $2^3S_1$ population, and D26 predicts a large optical depth there, placing M17 in the regime where profile distortion should be unambiguous. The SW bar is an ionization front viewed nearly edge-on ([21], hereafter P07), so sight lines through it cross long path lengths of ionized gas. Archival radio continuum imaging of the region is available (MAGPIS; [11]), and comparison with Hα imaging (VPHAS+; [8]), which unlike the radio is attenuated by dust, shows where extinction is strong.

In this paper, we report scattering-induced distortion of the He I λ10830 profile in M17 and test whether static models of the kind computed by D26 can account for it. Section 2 describes the observations and the data reduction. Section 3 presents the observed profiles and the evidence that their structure is produced by resonant scattering. Section 4 sets out the static models and the shift-invariant metrics used to compare them with the data, and shows that they fail at two of the pointings farthest from CEN 1. Sections 5 and 6 discuss the result and summarize. Wavelengths are given in vacuum, although line labels such as He I λ10830 follow the common practice of air wavelengths.

Observations

The observations were conducted on 2026 June 26 UT, using the WINERED echelle spectrograph ([14]) mounted on the 6.5 m Magellan Clay Telescope at Las Campanas Observatory. WINERED was operated in the HIRES-Y mode with the 0″.3 × 9″ slit, providing a spectral resolving power of $R \approx70{,}000$ and covering a wavelength range of 0.96–1.11 μm. Observations were carried out at six pointings, P1–P6, placed along a line extending from the ionizing source CEN 1 at a position angle (PA) of 229.4° (Figure 1a); coordinates and exposures are summarized in Table 1. WINERED has no image derotator, so the slit PA differs between pointings (Table 1). The pointings were observed in order from P6 to P1. Sky frames were obtained about 2°

ID	$\alpha$ (J2000)	$\delta$ (J2000)	$d_{\mathrm{CEN1}}$	UT (start)	$t_{\mathrm{exp}}$	Slit PA	$S/N_{\mathrm{peak}}$	$v_{\mathrm{LSR}}(\mathrm{Pa}\gamma)$
P1	18:20:28.52	$-16$:11:01.9	25	03:08	$60 \times3$	83	221	29.2
P2	18:20:26.95	$-16$:11:21.7	55	03:02	$60 \times3$	86	315	26.0
P3	18:20:25.09	$-16$:11:40.1	88	02:45	$60 \times3$	92	12	15.9
P4	18:20:24.37	$-16$:11:53.6	104	02:38	$60 \times3$	94	56	9.4
P5	18:20:23.01	$-16$:12:11.0	130	02:31	$60 \times3$	96	87	14.1
P6	18:20:21.89	$-16$:12:25.3	152	02:06	$60 \times6$	103	23	25.1

from M17 in Galactic latitude, three at a time, before and after P6, after P3, and after P1. The seeing measured by the DIMM was 0.′′6 to 0.′′8. Each target was observed at three dither positions along the slit, offset by +1′′, −1′′, and 0′′.

The data were reduced with the WARP pipeline v1.1 [9], which performs flat-fielding with exposures of a lamp-illuminated screen, rectification of the echelle orders, and wavelength calibration. The sky was subtracted from the raw frames beforehand, because the nebula fills the slit and the OH airglow doublet at $\lambda\lambda10834.241, 10834.338$ Å [20] overlaps the He I $\lambda10830$ profile. We scaled the six raw sky frames taken before and after each target so that the uncontaminated OH doublet at $\lambda\lambda10844.514, 10844.871$ Å matched its intensity in the target frame, and subtracted them before the reduction. Because the scaled sky frames also contain components that do not vary in proportion to the airglow, mainly persistence from a star observed before the first sky frames, this leaves a residual in the continuum, of up to 8% of the line peak (at P6), which we treat as described in § 4.2.

The rectified, wavelength-calibrated two-dimensional spectra of each pointing were shifted along the slit by the dither offsets and median-combined, and one-dimensional spectra were extracted over the central 40 pixels ($\sim6$′′) of the slit, rejecting pixels that depart from the spatial median by more than $4\sigma$ and rescaling the sum to the full aperture. For P1 alone, where the profile varies significantly along the slit, we additionally extracted spectra in eight bins of 5 pixels each ($\sim0.''75$, comparable to the seeing).

Dividing by the flat field removes most of the echelle blaze function. To measure what remains, an F-type star was observed immediately after the science exposures. It shows no significant telluric absorption at the He I $\lambda10830$ and Pa$\gamma$ wavelengths, so no telluric correction was applied. The residual sensitivity, derived with the star’s intrinsic Pa$\gamma$ absorption masked and interpolated over, departs from a constant by less than 0.1% across a line profile and by 1.3% between the two lines, which is immaterial to the profile shapes. The spectra are presented in counts, without absolute calibration.

Velocities are referred to a rest wavelength of 10941.082 Å for Pa$\gamma$ and, for He I $\lambda10830$, to the $gf$-weighted multiplet wavelength of 10833.138 Å, following D26. They are given in the local standard of rest (LSR) frame throughout, obtained from the wavelength solution with barycentric and solar-motion corrections applied, the latter adopting the standard solar motion of 20 km s$^{-1}$ toward $(\alpha,\delta)=(18^{\mathrm h},+30^\circ)$ in B1900. Within each pointing, Pa$\gamma$ also serves as a local velocity reference. Because its profile is asymmetric, we characterize its position by the profile bisector, the midpoint between the blue and red sides at the 50% level; its LSR velocity, $v_{\mathrm{LSR}}(\mathrm{Pa}\gamma)$, is listed in Table 1. Velocity intervals that are tied to the line in the measurements below are specified relative to this bisector. The profile metrics of § 4.2 are invariant under a rigid shift and depend only weakly on the reference, through the placement of the $\pm150\ \mathrm{km\ s^{-1}}$ window.

The Observed He I Profiles

Figure 2 shows the He I and Pa$\gamma$ emission-line profiles obtained at the six pointings. Every He I profile is multi-peaked: at least two peaks are resolved at each position, three at P2, P4, P5, and P6, and more at P1. At P5, troughs lie near the wavelengths of both the $J = 0$ and $J = 1, 2$ components at $v_{\mathrm{LSR}}(\mathrm{Pa}\gamma)$ (Figure 2), indicating that each component is split, as (D26) predicted for high $2\,^{3}\mathrm{S}_{1}$ column densities. The Pa$\gamma$ profiles, observed simultaneously through the same slit and reduced identically, remain single-peaked at every position. The lower level of Pa$\gamma$ ($n = 3$) is so sparsely populated that the

optical depth of the line is negligible,⁶ whereas that of He I $\lambda10830$ in M17 is of order $10^{2}$–$10^{3}$ (D26); Pa$\gamma$ therefore does not undergo resonant scattering and traces the velocity field of the gas almost directly. Moreover, hydrogen is ionized wherever helium is, so Pa$\gamma$ samples all of the gas that emits He I. The absence of corresponding peaks in Pa$\gamma$ therefore argues against a purely kinematic origin for the He I peaks. The shared exposures and reduction likewise rule out an instrumental or calibration origin.

Along the slit, the distortion changes little at most pointings but markedly at P1. Across eight independent bins there, each comparable to the seeing, the shape of the He I profile changes markedly, while the Paγ profile extracted from the same bins is nearly unchanged (Figure 3). Because Paγ is emitted along the same sight lines in each bin and is nearly unchanged, this variation cannot be explained by kinematics alone, and it is further evidence for resonant scattering. P1, however, lies on the steep flank of a compact Hα ridge (Figure 1a), and its sight lines evidently cross a complex gas structure; we leave its physical interpretation to future work.

To our knowledge, earlier observations of the line ([22, 15]) did not resolve the splitting of each multiplet component; the present observations resolve the splitting at P5 and, with

an optically thin comparison line observed in the same exposures, show that it arises from resonant scattering.

Beyond these common features, the profiles differ from sight line to sight line, in shape and in strength. At P2, the profile is dominated by a single bright peak with an asymmetric blue wing and two weak peaks blueward of it; Paγ shows a smooth tail extending to $v_{\mathrm{LSR}} \approx-30\ \mathrm{km\ s^{-1}}$; at this pointing, the scattered and kinematic contributions cannot be separated with the present data. In strength, the He I line at P3 and P4 is weak relative to the radio continuum, as is Hα (Figure 1b), consistent with the heavy obscuration of the SW bar ((P07)). Obscuration lowers the signal-to-noise ratio, which is lowest at P3 (Table 1), but being gray across a line profile, it can neither generate nor remove the structure discussed above.

The Paγ velocities themselves differ between sight lines. $v_{\mathrm{LSR}}(\mathrm{Pa}\gamma)$ ranges from +9.4 to +29.2 km s$^{-1}$ (Table 1), all redshifted, consistent with the recession of M17 as a whole ((P07)). The pattern is not monotonic with projected distance from CEN 1; together with the asymmetry of the Paγ profiles, it shows that more than one kinematic component lies along at least some sight lines.

Comparison with Static Models

The splitting of the profile confirms the prediction of D26 qualitatively. That prediction was made for a static, uniform sphere, and we now ask whether such models reproduce the observed profile shapes quantitatively.

The Model Grid

D26 treats resonant scattering in a static, spherically symmetric, uniform medium by Monte Carlo radiative transfer, and we use the emergent spectra made publicly available with that work ([7]). Following D26, throughout this paper, $\tau_{\mathrm{tot}}$ denotes the sum of the line-center optical depths of the $J=0$, 1, and 2 components, and $b$ is the Doppler parameter of the scattering atoms. We use the dust-free models with $b=6.45$, 10, 15, and 20 km s$^{-1}$, which cover, at each $b$, 33 values of $\tau_{\mathrm{tot}}$ from 0.01 to $10^6$, so the grid extends from the optically thin limit to optical depths far beyond the $\tau_{\mathrm{tot}}=2000$ that D26 estimates for the radio continuum peak of the SW bar (M17-B in D26), which corresponds to P3. Before comparison with the data, the model spectra are convolved with a Gaussian matched to the spectral resolution, $R=70{,}000$. We refer to the tabulated models as the grid and to the continuous set obtained by interpolating between them as the static family.

Shift-invariant Profile Metrics

Comparing the static grid with the observations requires a measure of profile shape, and the obvious choice is unavailable here: D26 proposed the mean blueshift as an observational diagnostic, but the asymmetry of the observed Pa$\gamma$ profile makes the systemic velocity difficult to fix, and any measure referred to it inherits that uncertainty. We therefore characterize each profile by its cumulative flux distribution, which can be described without a velocity zero point. Writing $v_q$ for the velocity at which the cumulative flux reaches $q\%$ of the total, we use two widths,

the first sensitive to the wings of the profile and the second to its core, together with the asymmetry parameter of G. Liu et al. [18],

which is negative when the median lies redward of the midpoint of the $v_{10}$–$v_{90}$ interval, that is, when the flux is concentrated to the red with a tail to the blue (a red-weighted profile); it vanishes for a symmetric profile and is bounded by $\pm1$. All three quantities are invariant under a rigid velocity shift.

Identical estimators are applied to the models and to the data: the spectrum is rebinned to $4\ \mathrm{km\ s^{-1}}$, negative values, which arise from noise, are set to zero so that the cumulative flux rises monotonically, and the flux is accumulated over $\pm150\ \mathrm{km\ s^{-1}}$ about $v_{\mathrm{LSR}}(\mathrm{Pa}\gamma)$ for the data and about the multiplet centroid for the models. A constant pedestal is subtracted from each observed spectrum, estimated from the line wings $100$–$150\ \mathrm{km\ s^{-1}}$ on either side of $v_{\mathrm{LSR}}(\mathrm{Pa}\gamma)$; at P1, the blue wing alone is used, the red wing of He I there containing a component with no counterpart in Pa$\gamma$ or at the other pointings. The interval $39 < v_{\mathrm{LSR}} < 54\ \mathrm{km\ s^{-1}}$, affected by residuals of the OH airglow, is not interpolated over, because doing so in the models would replace a narrow peak at its edge by a linear ramp that mimics a red wing. An OH residual of $\pm10\%$ of the doublet changes $D$ by at most $15\%$ at P4 and P5. At P3, the residual exceeds the line itself, and the comparison is not made. Because setting negative values to zero biases the metrics of faint lines, the models are compared as they would be observed: each model spectrum, scaled to the observed line flux, is given the noise of the observation and processed identically, and $\mathbf{m}_{s}$ below is the mean of 200 such realizations. Because the systemic velocity is uncertain, the models are also shifted by up to $\pm8\ \mathrm{km\ s^{-1}}$, and the smallest $D$ is adopted.

Uncertainties are obtained from 1000 realizations of each spectrum with Gaussian noise added. The noise variance in a pixel of flux $f$ is modeled as $a+bf$, with $a$ and $b$ fitted to the scatter among the three dither exposures; the noise is then scaled by 1.5 to account for the correlation between adjacent pixels. The three metrics differ in units and in precision, and because they are built from the same quantiles, their errors are correlated; the realizations therefore give the full covariance matrix $\mathbf{C}$ of $(W_{80},W_{50},A)$. The separation of an observed point $\mathbf{m}$ from the static family is measured by the Mahalanobis distance,

where $\mathbf{m}_{s}(b,\tau_{\mathrm{tot}})$ are the metrics of the static family; $D$ is the separation from the nearest static model in units of the measurement error in that direction. For a point drawn from a single model, $D^{2}$ follows a $\chi^{2}$ distribution with three degrees of freedom, and we quote the equivalent Gaussian significance on that basis. This is conservative, because minimizing over a two-parameter family leaves fewer degrees of freedom. Scaling the noise by 1.65 instead of 1.5 lowers the significances by at most $1.8\sigma$.

Results

Static models produce asymmetric profiles of either sign (see Figure 6b of D26, for $b=6.45\ \mathrm{km\ s^{-1}}$ and $\tau_{\mathrm{tot}}\leq10^{4}$). At low optical depth, the $J=0$ component, $34\ \mathrm{km\ s^{-1}}$ blueward of the $J=1,2$ pair, makes $A$ negative; as the optical depth rises, scattering moves weight into $J=0$, raising $A$, while both widths grow. For $W_{80}=55$–$70\ \mathrm{km\ s^{-1}}$, the static family thus has $A\gtrsim-0.13$ (Figure 4); only at $\tau_{\mathrm{tot}}\gtrsim2\times10^{5}$ does $A$ fall again, to about $-0.7$, with $W_{50}$ about $75\ \mathrm{km\ s^{-1}}$.

P4 and P5 lie outside this family, on its red-weighted side. They have $A=-0.27\pm0.03$ and $-0.29\pm0.01$ at $W_{80}=63$ and $58\ \mathrm{km\ s^{-1}}$, far more red-weighted than any static model of similar width; their separations are $D=12$ and $15$ ($12\sigma$ and $15\sigma$). P6, one of the two faintest lines, lies within the family ($D=1.0$), although, as at P5, its reddest peak is by far the strongest, unlike in the nearest models; the metrics are insensitive to this difference. P3, the faintest, is not compared, because residuals of the OH airglow exceed the line in the airglow interval (§ 4.2).

Figure 4c shows what the separation means for the profile at P5. The three nearest models of the grid lie at similar separations ($D=17$–18) along the degeneracy between $b$ and $\tau_{\mathrm{tot}}$. In each, however, the two strongest peaks are of nearly equal height or merge into a flat top, whereas in the observed profile, the reddest peak is clearly the stronger.

P1 and P2 are treated separately. At P1, the profile changes within the slit (§ 3), so the integrated spectrum combines sight lines of different character and does not test a single static medium; no static model with a core as broad as that at P1 ($W_{50}=33$ km s$^{-1}$) has wings as broad ($W_{80}=93$ km s$^{-1}$, against at most 64 km s$^{-1}$).

P2 lies near the static family, close to the models with $b=6.45$ km s$^{-1}$ and $\tau_{\rm tot}$ of a few ($D=2.4$). Its main peak falls in the airglow interval, and an OH residual of $\pm10\%$ of the doublet would change $D$ to between 0.7 and 4.5. Its profile, however, is not reproduced by those models. They have a distinct peak 34 km s$^{-1}$ blueward of the main peak, from the $J=0$ component, and none beyond it. In the observed profile, the $J=0$ component is not resolved from an asymmetric blue wing of the main peak, and two weak peaks lie 50 and 90 km s$^{-1}$ blueward of it, at 5% and 2.5% of its height. The metrics, being integrals over the profile, are insensitive to structure this weak, so their agreement with models of low optical depth does not establish that the line at P2 is unaffected by resonant scattering.

Discussion

The evidence presented here rests on profile shape rather than on line intensity, so it requires no absolute flux calibration and is not degenerate with extinction or

with the extent of the He$^{+}$ zone. The shape, however, is not that of a static, uniform sphere of any optical depth (§ 4), so at least one assumption of those models fails in M17. D26 noted that such deviations would not be surprising, because the stars and gas of real H II regions may be distributed in a complex way and gas flows may produce velocity gradients.

Of these possibilities, the data favor a velocity gradient within the scattering gas. A uniform motion cannot explain the departure, because it would shift the profile without changing its shape or the metrics. Both P4 and P5 have more negative $A$ than static models of comparable width, i.e., their flux is more strongly weighted to the red. Such red-weighted profiles are expected when resonance-line photons scatter in expanding gas, as A. H. Vaughan [22] argued for this line and as found for Ly$\alpha$ in outflows (A. Verhamme et al. [23]), and our preliminary radiative transfer calculations for an expanding shell confirm that $A$ then becomes more negative. Finally, the ionized layer of M17 is independently inferred to expand relative to the neutral shell and the molecular cloud ((P07)). We present the calculations, with a fit to the profile at P5 and the optical depth it implies, in a second paper.

Other departures from the models are not excluded. The comparison treats each profile as the emission of a single scattering medium, whereas more than one kinematic component lies along at least some sight lines (§ 3); a superposition of static components at different velocities could alter the metrics, although it would have to produce in He I the peaks that are absent from Pa$\gamma$. The models are spherical and uniform, and their spectra are integrated over the whole sphere, whereas each observation samples a single line of sight.

The optical depth of the $2\,{}^{3}\mathrm{S}_{1}$ level matters beyond the He I $\lambda10830$ line itself. Determinations of the primordial helium abundance ($Y_{\mathrm{p}}$) from He I recombination lines correct the line intensities for this optical depth, which is not observed but fitted together with the density and temperature ([16, 2, 3]), using radiative transfer corrections computed for a spherical nebula without systematic velocity gradients ([4, 17]). The He I $\lambda10830$ profile is shaped by this optical depth directly, and at two of the six pointings, models without velocity gradients fail to reproduce it. Resolved profiles of this line therefore offer an independent check on those corrections.

Summary

1. We resolve the He I $\lambda10830$ profile in the H II region M17 into the multiple peaks expected from resonant scattering. At six pointings in M17 (P1 to P6, in order of distance from CEN 1), the He I profiles are multi-peaked while Pa$\gamma$, observed in the same exposures, is single-peaked.

2. Static scattering models do not account for the profiles at P4 and P5, two of the four pointings farthest from CEN 1. In the shift-invariant metrics ($W_{80}$, $W_{50}$, $A$), these lie outside the static family of D26 by $12\sigma$ and $15\sigma$. At P6, the metrics do not distinguish the profile from static models, and at P3, residuals of the OH airglow preclude the comparison.

3. We argue that the discrepancy reflects a velocity gradient within the scattering gas, because both P4 and P5 are more red-weighted than static models of comparable width, as expected for scattering in expanding gas. The ionized layer of M17 is independently inferred to expand. The He I $\lambda10830$ profile is thus shaped by both the kinematics and the optical depth of the He$^{+}$ zone, and resolved profiles could provide a new diagnostic of these properties in H II regions.

Acknowledgments

We thank B. T. Draine for making the model spectra of D26 publicly available. We also thank the staff of Las Campanas Observatory for their support during the observations, and the staff of Carnegie Observatories, especially Andy McWilliam and Shreyas Vissapragada, for enabling the operation of WINERED.

This paper includes data gathered with the 6.5 meter Magellan Telescopes located at Las Campanas Observatory, Chile. WINERED was developed by the University of Tokyo and the Laboratory of Infrared High-resolution spectroscopy (LiH), Kyoto Sangyo University, with financial support from JSPS KAKENHI (grant Nos. 16684001, 23040042, 21840052, and 26287028), and from the MEXT Supported Program for the Strategic Research Foundation at Private Universities (Nos. S0801061 and S1411028).

The National Radio Astronomy Observatory is a facility of the National Science Foundation operated under cooperative agreement by Associated Universities, Inc. Based on data products from observations made with ESO Telescopes at the La Silla Paranal Observatory under programme ID 177.D-3023, as part of the VST Photometric H$\alpha$ Survey of the Southern Galactic Plane and Bulge (VPHAS+, www.vphas.eu). This research has made use of NASA’s Astrophysics Data System.

H.S. is supported by JSPS KAKENHI Grant Numbers JP22K03683, JP19K03917. The 2026 June observing run was partly supported by the TMU Research Fund for Young Scientists from Tokyo Metropolitan University.

During the preparation of this work, the authors used Claude (Anthropic) in planning the observations, in writing analysis and plotting code, and in drafting and editing the text. The authors verified the code and its results, reviewed and edited the text, and take full responsibility for the content of the publication.

Author Contributions

H.S. planned the observations and supervised them remotely, carried out the data analysis, and wrote the manuscript. D.T. and N.M. contributed to the observing proposal and the interpretation of the results. D.T., S.O., and T.T. operated the instrument at the telescope. All authors discussed the results and commented on the manuscript.

Facilities: Magellan:Clay (WINERED), VLA, Ef-selsberg, ESO:VST (OmegaCAM).

Software: WARP v1.1, astropy ([1]), numpy ([10]), scipy ([24]), matplotlib ([13]), seaborn ([25]), Claude (Anthropic).