betaFeedback

WorksPhysics and Astronomy

Wolf-Rayet Stars in M33: The Impact of Stellar Crowding on Spectral Classification and Mass-Loss Diagnostics

Abstract

The metallicity dependence of Wolf-Rayet (WR) mass-loss rates is a critical input for massive star evolutionary models, chemical yield predictions, and ionizing radiation budgets of young stellar populations. The Triangulum Galaxy (M33), with 213 spectroscopically confirmed WR stars spanning a ∼\sim0.5 dex radial metallicity gradient, is the premier laboratory for this calibration. However, the ground-based fiber spectra on which the current census rests blend each target with neighboring OB stars, diluting diagnostic emission-line equivalent widths and biasing spectral classifications, inferred mass-loss rates, and surface hydrogen abundances. Using 0\farcs1-resolution HST/PHATTER survey imaging, we quantify this contamination for 123 WR stars within the survey footprint, finding that ∼\sim27% suffer moderate-to-severe dilution from unresolved neighbors with WR continuum flux contributions FWR<0.85F_{\rm WR} < 0.85. We provide crowding classifications, FWRF_{\rm WR} fractions, and six-band photometry in the F275W, F336W, F475W, F814W, F110W, and F160W bands for each identified counterpart. We also present Keck/DEIMOS slit spectra of five WR stars, which are substantially less contaminated than the discovery MMT/Hectospec fiber spectra. A direct equivalent-width comparison of these targets confirms the predicted dilution. The slit spectra yield blue-edge velocities of He I absorption (vblue≈700v_{\rm blue} \approx 700–15001500 km s−1^{-1}), comparable to WR stars at similar metallicity in the LMC. The FWRF_{\rm WR} correction factors, together with the slit-spectrum benchmarks, provide the inputs needed to place M33 on the M˙\dot{M}–ZZ relation as a calibration anchor between the LMC and the Galaxy.

Introduction

Massive stars have brief yet influential lifetimes, and significantly impact the interstellar medium (ISM) through powerful stellar winds, radiation, and their ultimate fate as core-collapse supernovae [28, 51, 10]. Wolf-Rayet (WR) stars represent a particularly consequential evolutionary phase, characterized by mass-loss rates spanning M˙∼10−6\dot{M} \sim10^{-6}–10−4 M⊙ yr−110^{-4}\ M_{\odot}\ \mathrm{yr}^{-1} and helium-rich surfaces that significantly shape galactic chemical enrichment [8, 10, 33, 19, 22]. At approximately Galactic and Large Magellanic Cloud (LMC) metallicities, WR stars may originate from O-type progenitors with initial masses above ∼20\sim20–25 M⊙25\ M_{\odot} [46] and occupy the WR phase for roughly the final 10% of their lifetimes [33]. They serve as markers of recent high-mass star formation and are potential progenitors of stellar-mass black holes and long-duration gamma-ray bursts [56, 25]. The defining observational characteristic of WR stars is their broad emission-line-dominated spectra, produced by dense, fast stellar winds [10]. Two primary channels produce WR-type spectra: very massive main-sequence stars (≳80\gtrsim80–100 M⊙100\ M_{\odot}) that retain

Email: [email protected] hydrogen-rich envelopes (WNh stars), and classical WR (cWR) stars—core helium-burning, hydrogen-depleted objects with Teff>30,000T_{\mathrm{eff}} > 30{,}000 K that constitute ∼90%\sim90\% of the known population ([18, 48, 45]).

Observations of WR stars provide critical constraints on massive star evolutionary models, particularly the poorly understood interplay between mass loss, rotation, and metallicity ([7, 51] (Ou et al. 2023)). In the Milky Way, non-LTE atmospheric analyses with codes such as CMFGEN ([24]) and PoWR ([21]) have yielded relatively precise wind parameters for individual WR stars ([22, 43]), but the narrow metallicity range within the Galaxy limits tests of the M˙\dot{M}–ZZ dependence. Surveys in the Magellanic Clouds have extended this baseline to sub-solar metallicities ([20, 47]), and population synthesis studies incorporating massive binaries ([13]) have demonstrated the importance of binary channels. However, fully constraining the metallicity dependent changes requires measuring the characteristics of WR populations spanning a broad range of chemical environments, both within galaxies with known abundance gradients and across samples of galaxies at different mean metallicities.

The nearby Triangulum galaxy (M33) provides such a laboratory. At a distance of ∼840\sim840 kpc ([16, 27]), it is close enough for individual massive stars to be resolved as well as a single survey can capture a statistically significant WR population spanning a range of galactic environments. Its relatively face-on disk orientation (i≈56∘i \approx56^\circ; [41]) minimizes differential extinction along the line of sight, while its high star-formation rate and large luminosity ensure a rich massive star population ([35, 11]). Most critically, M33 hosts a well-characterized radial metallicity gradient spanning ∼0.5\sim0.5 dex from the nucleus to the outer disk ([42, 5]), enabling a direct examination of how chemical environment shapes WR star properties such as wind velocities and mass-loss rates ([6, 49, 40]). The spectroscopic census by Massey [28], Neugent & Massey [35, 36], and Massey et al. [30]—conducted primarily with the fiber-fed Hectospec spectrograph on the 6.5 m Multiple Mirror Telescope (MMT) ([15])—has identified 213 WR stars in M33, establishing the subtype distribution and spatial demographics across the disk.

In the course of the Triangulum Extended (TREX) spectroscopic survey of M33 ([17, 39, 11]), we used the Keck/DEIMOS spectrograph to obtain spectra of ≈7000\approx7000 resolved stars across the galaxy’s disk and halo. TREX targets in M33’s inner disk were preselected from the HST\mathit{HST} Panchromatic Hubble Andromeda Treasury: Triangulum Extended Region (PHATTER) imaging survey ([55]) to be relatively isolated point sources. During our systematic inspection of these spectra, we found five previously known WR stars as serendipitous detections. Because the targets were selected for isolation and observed in sub-arcsecond seeing (\lesssim0\farcs8) through 0\farcs8–1\farcs0-wide slits, these slit spectra are substantially less contaminated by neighboring sources than the original Hectospec fiber spectra (1\farcs5 diameter; observed under typical \sim1\arcsec seeing at the MMT; [31]) from which the WR classifications were derived.

This serendipitous discovery, and the visibly cleaner spectral features it revealed, motivated us to carry out a systematic analysis of the broader WR population within the PHATTER footprint. At M33’s distance, a typical 1\arcsec seeing disk corresponds to ∼4\sim4 pc — comparable to the typical size of compact OB associations — such that the 1\farcs5 Hectospec fibers sometimes blend the WR target with nearby massive neighbors. These unresolved OB companions dilute diagnostic emission lines (e.g., He II λ4686\lambda4686, C IV λ5808\lambda5808) by contributing continuum flux, reducing measured equivalent widths (EWs) and potentially biasing spectral classifications and mass-loss rate determinations ([57, 4, 10]). An additional source of contamination is narrow nebular Hα\alpha emission from the diffuse ISM, which can be captured by fiber apertures without the spatial information needed to separate it from the broad stellar wind profile, artificially inflating the apparent Hα\alpha luminosity and hence the inferred M˙\dot{M} ([10]). As we demonstrate in § 4.2, long-slit spectrographs such as Keck/DEIMOS partially mitigate this by enabling spatial subtraction of the extended nebular component.

Using 0\farcs1-resolution HST\mathit{HST} imaging from the PHATTER survey ([55]), we quantify the flux contribution from neighboring sources for 123 of the 213 M33 WR stars within the survey footprint, classify their crowding environments, and provide robust six-band photometry. We also present multi-epoch Keck/DEIMOS spectra for three of the five serendipitously observed WR stars, spanning a 4–5 year baseline. Together, the crowding-corrected photometric catalog and the new slit spectra provide a vital reference for future observational programs, population synthesis calibrations, and the development of machine learning classifiers.

This paper is organized as follows. In § 2, we describe the observational datasets: the M33 WR catalog, the PHATTER imaging, and the Keck/DEIMOS spectroscopy. In § 3, we present the crowding analysis methodology and color-magnitude diagram results. In § 4, we analyze the serendipitous spectra, including nebular Hα\alpha decontamination and temporal variability. In § 5, we discuss implications for spectral classification, mass-loss rates, and population synthesis. In § 6, we summarize the main results.

Data

This study uses the published spectroscopic catalog of M33 WR stars (§ 2.1) to define the parent sample, together with two observational datasets: HSTHST/PHATTER imaging (§ 2.2) and Keck/DEIMOS slit spectroscopy from the TREX survey (§ 2.3).

The M33 Wolf-Rayet Parent Sample

The parent sample for our study is the comprehensive catalog of WR stars in M33, which is the result of several foundational spectroscopic surveys carried out using the 6.5-m MMT/Hectospec multi object fiber spectrograph [15]. The primary census by Neugent & Massey [35] listed 206 WR stars and an additional six candidates. Subsequent work by Neugent & Massey [36] confirmed five of these candidates as bona fide WR stars. Including another confirmation by Massey et al. [31] and one new discovery listed in Massey et al. [30], the total known population of spectroscopically identified WR stars in M33 currently stands at 213. We note that this census includes objects with complex classifications. For instance, the star J013418.37+303837.0 (Var 2) is a known Luminous Blue Variable (LBV), but it exhibits an Ofpe/WN9 spectrum and is often included in WR population counts [30]. The issue of stellar crowding for M33 WR stars has been previously considered, particularly in the dense cores of the NGC 604 and NGC 595 young clusters [29, 35], and surveyed using HST by Drissen et al. [12].

Hubble Space Telescope Photometry from the PHATTER Survey

The M33 photometry is drawn from the PHATTER survey [55], a 108-orbit HSTHST program that imaged ∼22\sim22 million resolved stars across the inner disk in six filters: F275W and F336W in the near ultraviolet (the latter is similar to ground based UU) using Wide Field Camera 3 ultraviolet/visible light (WFC3/UVIS), F475W and F814W in the optical (similar to ground based BB and II, respectively) using Advanced Camera for Surveys wide field (ACS/WFC), and F110W and F160W in the near infrared (similar to ground based JJ and HH, respectively) using WFC3 near-infrared (WFC3/IR). The survey footprint comprises three contiguous bricks covering an approximately 12′×18′12' \times18' region; full details of the tiling strategy, photometric pipeline, and forced PSF-fitting catalogs are given by Williams et al. [55]. The catalogs are publicly available via MAST.1

Of the 213 known M33 WR stars, 123 fall within the PHATTER footprint. These 123 targets form the basis of our crowding analysis (§ 3). The 0\farcs1-resolution PHATTER imaging enables the first systematic, star-by-star crowding characterization of this population.

Keck DEIMOS Spectroscopy

Spectroscopic data were obtained using the DEep Imaging Multi-Object Spectrograph (DEIMOS) mounted on the Keck II 10-meter telescope at the summit of Mauna Kea, Hawaii. DEIMOS, developed by Faber et al. [14], provides medium spectral resolution and was an integral part of the TREX, which targeted stars across M33’s disk and halo to explore its stellar kinematics [17, 39, 11]. A total of ≈7000\approx7000 stars were observed during the TREX survey, making it currently the largest stellar spectroscopic survey of M33, spanning the galaxy’s entire disk out to a deprojected radius of ∼11\sim11 kpc [17, 39, 11]. Of the 123 WR stars within the PHATTER footprint, five WR stars were captured by TREX as serendipitous detections (§ 4).

Spectroscopic targets were selected from the M33 HSTHST PHATTER survey. The DEIMOS observations were obtained using two gratings: (1) 600ZD (600 line mm−1^{-1}, R≈2000R \approx2000) for spectra targeting young blue stars, and (2) 1200G (1200 line mm−1^{-1}, R≈6000R \approx6000) for older, red giant branch (RGB) stars. The spectral coverage is λλ∼4600\lambda\lambda\sim4600–9800 Å and λλ∼6300\lambda\lambda\sim6300–9800 Å for the 600ZD and 1200G gratings, respectively.

Data were reduced with the spec2d and spec1d pipelines [9, 37], which perform flat-fielding, sky subtraction, wavelength calibration, and slit-centering corrections. Heliocentric radial velocities were measured by cross-correlating the observed spectra with stellar templates (Tonry & Davis [52]).

Complex Environment of Wolf-Rayet Stars

The primary goal of our analysis is to robustly identify the HST counterparts to the ground-based M33 WR catalog and to quantify the degree of photometric contamination affecting each source. As discussed in § 1, ground-based seeing blends WR targets with neighboring sources; our methodology quantifies this contamination using the PHATTER catalog.

Cross-Matching and Counterpart Identification

For each of the 123 targets, we queried the PHATTER point-source catalog (Williams et al. [55]) for all resolved sources within 1″ of the published ground-based WR coordinates, a radius comparable to the typical seeing disk of the foundational surveys (Neugent & Massey [35]). We identified the WR counterpart as the brightest F275W source within this aperture: classical WR stars (Teff≳30,000T_{\rm eff} \gtrsim30{,}000 K; Crowther [10]; Shenar [45]) emit predominantly in the ultraviolet, providing maximum contrast against cooler OB neighbors.

Using this criterion, we securely identified a unique, bright F275W counterpart for 113 of the 123 cases. The median F275W magnitude of the identified counterparts is 18.6 (interquartile range 17.0–19.2), well above the PHATTER 50% completeness limit (mF275W≈24.1m_{\rm F275W} \approx24.1–24.4 across all stellar density bins Williams et al. [55]). At this brightness the surface density of PHATTER sources (ρ≈1.4×10−2\rho\approx1.4 \times10^{-2} arcsec−2^{-2}) yields a chance superposition probability within 1\arcsec of only ∼4%\sim4\%. The remaining 10 targets possess only faint counterparts (mF275W≳23m_{\mathrm{F275W}} \gtrsim23). Because the chance alignment probability at these faint magnitudes rises above ∼40%\sim40\%, we flag these ∼10\sim10 targets as ambiguous and discuss them separately in § 3.5.

Synthetic Fiber Aperture Model

To simulate the ground-based spectroscopic observations from which the parent WR catalog was constructed, we developed a synthetic fiber aperture analysis. For each target, we centered a circular aperture of radius r_{\mathrm{fib}} = 0\farcs75 on the published WR coordinates, matching the 1\farcs5-diameter fibers of the Hectospec multi-object spectrograph on the 6.5 m MMT ([15]). We queried all PHATTER sources within 2\farcs87 (5σ5\sigma of the adopted seeing model; see below) of the catalog coordinates and modeled the ground-based flux distribution of each source as a circular two-dimensional Gaussian,

I(r)=I0exp⁡(−r22σ2),(1)I(r) = I_0 \exp\left(-\frac{r^2}{2\sigma^2}\right), \tag*{(1)}

where rr is the angular distance from the source centroid and σ=FWHM/(22ln⁡2)\sigma= \mathrm{FWHM}/(2\sqrt{2\ln2}). We adopted FWHM =

1\farcs0, representative of median ground-based seeing at the observatories used for the foundational surveys ([35, 30]). This Gaussian approximation is standard practice for crowding simulations in dense stellar fields ([34]). For each source ii at angular offset did_i from the fiber center, the fraction of its total flux captured by the fiber aperture is

fi=12πσ2∬x2+y2≤rfib2exp⁡(−(x−xi)2+(y−yi)22σ2) dx dy,(2)f_i = \frac{1}{2\pi\sigma^2}\iint_{x^2+y^2 \le r_{\mathrm{fib}}^2} \exp\left(-\frac{(x-x_i)^2+(y-y_i)^2}{2\sigma^2}\right)\,dx\,dy, \tag*{(2)}

where (xi,yi)(x_i,y_i) is the offset of source ii from the fiber center in arcseconds. This integral is evaluated numerically on a fine grid (0\farcs005 pixel−1^{-1}); for a centered source (di=0d_i = 0) it reduces analytically to 1−exp⁡(−rfib2/2σ2)1-\exp(-r_{\mathrm{fib}}^2/2\sigma^2), which we use as a sanity check. The fractional flux contribution of source ii to the total fiber-captured flux is then

Fi=fi⋅10−0.4mi∑jfj⋅10−0.4mj,(3)F_i = \frac{f_i\cdot10^{-0.4m_i}}{\sum_j f_j\cdot10^{-0.4m_j}}, \tag*{(3)}

where the sum runs over all PHATTER sources within the search radius and mim_i is the apparent magnitude of source ii in the relevant filter. For the identified WR counterpart, FWR≡FiF_{\rm WR} \equiv F_i provides a physically motivated, model-dependent measure of contamination that complements the model-independent crowding metric defined in § 3.3. We evaluated Equation (3) independently in the F275W, F336W, F475W, and F814W filters. FWRF_{\rm WR} decreases monotonically from UV to optical wavelengths (median values: 0.90, 0.88, 0.69, and 0.46, respectively), reflecting the increasing relative contribution of cooler neighbors at longer wavelengths. The difference ΔFWR≡FWR275−FWR814\Delta F_{\rm WR} \equiv F_{\rm WR}^{275} - F_{\rm WR}^{814} has a median of 0.28 with an interquartile range of 0.11–0.47 across our 123-star sample; 77% of targets show ΔFWR>0.10\Delta F_{\rm WR} > 0.10, confirming that the wavelength dependence is not negligible for the majority of the sample. This has two practical consequences. For crowding classification and counterpart identification, we adopt the F275W band, where the contrast between hot WR continua and cooler neighbors is greatest (§ 3.1). For quantitative EW corrections of optical spectral diagnostics, however, the appropriate FWRF_{\rm WR} is that evaluated at the wavelength of the emission line: the tabulated F475W-band value for each star applies to lines at ∼4500\sim4500–5500 Å (e.g., He II λ4686\lambda4686, He II λ5411\lambda5411), and the F814W-band value applies to lines at ≳7000\gtrsim7000 Å. Both filter-band FWRF_{\rm WR} values are provided in the machine-readable catalog.

The physical interpretation of FWRF_{\rm WR} as a contamination measure is supported by the residual astrometry. The angular offset dd between the identified PHATTER counterpart and the published catalog coordinates serves as a model-independent crowding diagnostic (§ 3.3): for an isolated WR star the ground-based flux-weighted centroid coincides with the true stellar position (d≈0d \approx0), whereas blending with a nearby comparably-bright neighbor pulls the centroid toward the secondary, increasing dd. Figure 2 quantifies this effect. After applying a bulk astrometric correction to place the ground-based catalog on the HST/PHATTER reference frame (§ 3.1), the residual positional offsets show a clear dependence on crowding class: isolated WR stars have a median residual of 0. ⁣ ⁣′′130.\!\!^{\prime\prime}13 (σ≈0. ⁣ ⁣′′10\sigma\approx0.\!\!^{\prime\prime}10), consistent with the intrinsic astrometric precision of the ground-based surveys ([35]), while moderately and very crowded sources show systematically larger residuals—medians of 0. ⁣ ⁣′′220.\!\!^{\prime\prime}22 and 0. ⁣ ⁣′′260.\!\!^{\prime\prime}26, respectively, with scatter increasing to σ≈0. ⁣ ⁣′′19\sigma\approx0.\!\!^{\prime\prime}19–0. ⁣ ⁣′′330.\!\!^{\prime\prime}33—consistent with flux-weighted centroid displacement toward unresolved bright neighbors. The maximum offset among very crowded stars reaches 0. ⁣ ⁣′′880.\!\!^{\prime\prime}88.

Astrometric residuals between ground-based catalog positions and HST/PHATTER counterparts for 123 M33 WR stars

Figure 2. Astrometric residuals between ground-based catalog positions and HSTHST/PHATTER counterparts for 123 M33 WR stars within the PHATTER footprint, color-coded by crowding classification. A bulk systematic offset between the ground-based astrometric frame and the HSTHST/PHATTER reference frame has been removed prior to this comparison (§ 3.1). (a) Vector-point diagram showing Δαcos⁡δ\Delta\alpha\cos\delta vs. Δδ\Delta\delta, the residual offset components in right ascension and declination. Reference circles at 0\farcs10, 0\farcs25, and 0\farcs50 are shown for scale. Green circles, orange squares, and red diamonds denote the isolated (N=90N = 90), moderately crowded (N=13N = 13), and very crowded (N=20N = 20) samples, respectively. Isolated sources cluster tightly around the origin, while very crowded sources extend to larger, more isotropically distributed offsets. (b) Histogram of the total angular offset ∣Δθ∣|\Delta\theta| for each crowding class; vertical lines mark the class medians (isolated: 0\farcs13; moderately crowded: 0\farcs22; very crowded: 0\farcs26, maximum 0\farcs88). The progressive broadening of the offset distribution from isolated to very crowded is a direct, model-independent signature of ground-based centroid displacement by unresolved blended neighbors, consistent with the crowding classifications derived from the two-parameter photometric metric (Figure 3).

The Crowding Metric

While the fractional flux contribution FWRF_{\rm WR} (Equation (3)) provides a physically motivated measure of contamination, it requires adopting a specific seeing model and fiber geometry. As a complementary, model-independent characterization of source confusion, we developed a two-parameter crowding diagnostic based on quantities measured directly from the PHATTER catalog.

For each WR target, we define:

  • dd: the angular offset between the published ground-based WR coordinates and the identified PHATTER counterpart (§ 3.1);

  • Δm12\Delta m_{12}: the F275W magnitude difference between the brightest and second-brightest PHATTER sources within 1″ of the catalog coordinates.

These parameters provide complementary indicators of local source confusion. A larger positional offset dd indicates a greater discrepancy between the published

This table is published in its entirety (213 entries) in machine-readable format. A portion is shown here for guidance regarding its form and content.

ground-based position and the identified PHATTER counterpart, while a smaller Δm12\Delta m_{12} indicates the presence of a comparably bright nearby source. We therefore use their joint distribution to characterize the degree of local crowding.

The (d,Δm12)(d,\Delta m_{12}) metric is intended as an empirical, model-independent description of source confusion rather than as a quantitative estimate of the fiber contamination. Larger positional offsets and smaller magnitude contrasts both indicate greater potential confusion, motivating a two-dimensional partition of the observed distribution. We therefore adopt simple linear boundaries in the (d,Δm12)(d,\Delta m_{12}) plane to define qualitative crowding categories. Quantitative contamination estimates are instead based on the independently calculated per-star FWRF_{\mathrm{WR}} values from the full synthetic-fiber model (§ 3.2).

The (d,Δm12)(d,\Delta m_{12}) distribution (Figure 3) shows a concentration of sources at small dd and large Δm12\Delta m_{12} and a tail toward larger offsets and smaller magnitude contrasts. For a compact and reproducible description of this continuous distribution, we define three crowding categories using the following empirical boundaries:

Two-parameter crowding diagnostic plot showing magnitude difference and positional offset

Figure 3. Two-parameter crowding diagnostic for 123 M33 WR stars within the PHATTER footprint, evaluated in the F275W band. The horizontal axis shows the angular offset dd of the brightest F275W PHATTER source from the published ground-based WR coordinates; the vertical axis shows the magnitude difference Δm12\Delta m_{12} between the brightest and second-brightest F275W sources within 1″. The dashed and dotted lines show the empirical classification boundaries Δm12=8.2×d\Delta m_{12} = 8.2 \times d (isolated/moderately crowded) and Δm12=4.0×d\Delta m_{12} = 4.0 \times d (moderately crowded/very crowded), respectively (§ 3.3). Green circles, orange squares, and red diamonds mark the isolated, moderately crowded, and very crowded samples. Over one quarter of the sample (moderately + very crowded) has ground-based photometry and spectral classifications potentially compromised by unresolved continuum dilution.

Isolated:Δm12≥8.2×d,Moderate:4.0×d≤Δm12<8.2×d,Very Crowded:Δm12<4.0×d.\begin{aligned} \textbf{Isolated:}\quad&\Delta m_{12} \ge8.2 \times d,\\ \textbf{Moderate:}\quad&4.0 \times d \le\Delta m_{12} < 8.2 \times d,\\ \textbf{Very Crowded:}\quad&\Delta m_{12} < 4.0 \times d. \end{aligned}

These boundaries are descriptive rather than contours of constant FWRF_{\mathrm{WR}}. The independently computed FWRF_{\mathrm{WR}} values nevertheless show the expected broad ordering, with isolated sources generally having larger WR flux fractions and crowded sources smaller values, while exhibiting substantial scatter (Table 1). This scatter is expected because FWRF_{\mathrm{WR}} includes all PHATTER sources contributing to the synthetic fiber, whereas the (d,Δm12)(d,\Delta m_{12}) metric uses only the local positional offset and the two brightest sources within 1″. As a sensitivity check, varying both linear slope parameters by ±15%\pm15\% changes the classification of only four to five of the 123 targets, indicating that the descriptive sample divisions are insensitive to modest changes in the adopted boundaries.

IAU NameR.A.a^{a}Decl.a^{a}Cross-IDSpectral TypeCrowding Categorydbd^{b} (arcsec)Δm12c\Delta m_{12}^{c} (mag)FWRdF_{\mathrm{WR}}^{d}Coord. Source
M33 J013326.57+303550.123.36072430.597252⋯\cdotsWN6Isolated0.1283.660.82PHATTER
M33 J013333.26+304129.623.38859830.691548AM 5WC6Moderately Crowded0.1750.740.41PHATTER
M33 J013333.79+304129.723.39080030.691570N595-WR9WN7Very Crowded0.2490.120.32PHATTER
M33 J013340.18+303551.623.41741430.597660MC 40WC4-5Isolated0.1294.590.84PHATTER
M33 J013343.30+303533.923.43043330.592747UIT 177WN4.5+O6-9Moderately Crowded0.1170.470.46PHATTER
M33 J013348.81+303949.623.45336530.663783MJ X4WNIsolated0.1701.620.47PHATTER
M33 J013353.78+303528.523.47408030.591256MJ G34WNEVery Crowded0.1780.140.29PHATTER
M33 J013358.67+303525.723.49446030.590461OB 2-4B1 Ia+WNEModerately Crowded0.7624.100.73PHATTER
M33 J013404.03+304658.223.51678730.782825⋯\cdots⋯\cdotsIsolated0.1064.240.94PHATTER
M33 J013411.12+304636.923.54632230.776907⋯\cdotsWC4Very Crowded0.5530.590.34PHATTER

Table 1. Catalog of M33 Wolf-Rayet Stars with Crowding Analysis

The physical interpretation of each category is as follows:

  1. Isolated (N=90N = 90; 73%): The WR counterpart strongly dominates its immediate neighborhood: no comparably bright source lies within 1′′1^{\prime\prime}. For the majority of this subsample, FWR≳0.85F_{\mathrm{WR}} \gtrsim0.85 in the synthetic fiber model (§ 3.2), and spectral classifications are considered reliable. However, a small number of isolated targets in dense spiral-arm environments have lower FWRF_{\mathrm{WR}} values (down to ∼0.5\sim0.5) due to the cumulative contribution of many faint sources at 1′′1^{\prime\prime}–3′′3^{\prime\prime} offsets that fall outside the 1′′1^{\prime\prime} metric search radius but within the fiber aperture. For these objects, the counterpart identification is confident but the fiber-level EW dilution may still be significant.

  1. Moderately Crowded (N=13N = 13; 11%): A secondary source within 1′′1^{\prime\prime} is bright enough to have measurably displaced the ground-based centroid. FWRF_{\mathrm{WR}} typically falls in the range ∼0.40\sim0.40–0.850.85. The expected EW dilution of ∼20\sim20–50%50\% is sufficient to shift absolute-EW classifications by one subtype in some cases [50, 10]. The WR counterpart identification remains confident.

  1. Very Crowded (N=20N = 20; 16%): Multiple bright sources within 1′′1^{\prime\prime} contribute comparably to the fiber flux (FWR≲0.50F_{\mathrm{WR}} \lesssim0.50 in most cases). Observed EWs may be diluted by factors of ≳2\gtrsim2; spatially resolved spectroscopy is required for definitive characterization.

The (dd, Δm12\Delta m_{12}) classification and FWRF_{\mathrm{WR}} characterize crowding on complementary spatial scales: the former uses the counterpart positional offset and the magnitude contrast of the two brightest sources within 1′′1^{\prime\prime}, while FWRF_{\mathrm{WR}} includes all PHATTER sources contributing to the synthetic fiber. Both quantities are reported in Table 1. Crowded targets are preferentially located in the spiral arms and the cores of the giant H II regions NGC 604 and NGC 595 (Figure 1), where the surface density of massive stars is highest, as expected. We note that the crowding metric evaluated in F475W yields broadly consistent classifications (Isolated: 87; Moderate: 15; Crowded: 21). The small differences arise because cooler companions contribute relatively more flux at longer wavelengths. We adopt the F275W classification as our preferred scheme throughout, consistent with the counterpart identification in § 3.1.

Spatial distribution of the Wolf-Rayet population in M33

Figure 1. Spatial distribution of the Wolf-Rayet population in M33. (a) Full-galaxy view in a GALEX near-ultraviolet (NUV) image. Prominent star formation is visible along the spiral arms. Open white circles indicate the positions of all 213 spectroscopically confirmed WR stars from Massey [28], Neugent & Massey [35, 36], and Massey et al. [30]. The cyan outline indicates the HST/PHATTER survey footprint Williams et al. [55], which covers the inner high-surface-brightness disk and encloses 123 of the 213 known WR stars. Thick cyan lines denote the boundaries of the three major survey bricks. The giant H II regions NGC 604 and NGC 595 are labeled. (b) Zoomed in view of the PHATTER footprint using the HST/WFC3 F336W mosaic as background, with WR stars color-coded by crowding classification: isolated (green circles; N=90N = 90), moderately crowded (orange squares; N=13N = 13), and very crowded (red diamonds; N=20N = 20), as defined in § 3.3. Blue stars mark the five WR targets with Keck/DEIMOS spectra from the TREX survey (§ 4). The very crowded WR stars cluster preferentially along the spiral arm and in the vicinity of NGC 604 and NGC 595, confirming that the photometric contamination quantified in this work is a systematic effect driven by the dense stellar environments associated with active star formation.

Color-Magnitude Diagram Analysis

We constructed global and local CMDs to place the WR stars in evolutionary context and to provide an independent check on the crowding classification. Figure 4 shows the WR sample overplotted on the full PHATTER Hess diagram, color-coded by crowding category. The isolated sample (green) occupies the expected WR locus—luminous and extremely blue (F336W–F475W ≲−0.3\lesssim-0.3), consistent with Teff≳30,000T_{\mathrm{eff}} \gtrsim30{,}000 K [10]. The moderately and very crowded samples largely share this locus, confirming that the F275W-selected counterparts are genuine hot sources rather than photometric artifacts.

Global color-magnitude diagrams of M33 with the Wolf-Rayet sample overplotted

Figure 4. Global color-magnitude diagrams (CMDs) of M33 from the PHATTER survey with the WR sample overplotted, color-coded by crowding classification. (a) NUV CMD: F336W vs. F275W−F336W. (b) Optical CMD: F475W vs. F336W−F475W. The greyscale Hess diagram shows the density of the full PHATTER source catalog (∼22\sim22 million stars). Green circles, orange squares, and red diamonds denote the isolated (N=90N = 90), moderately crowded (N=13N = 13), and very crowded (N=20N = 20) WR samples, respectively (§ 3.3). Blue stars mark the five WR targets with Keck/DEIMOS spectra from the TREX survey (§ 4). Three WR stars lacking valid F336W photometry (two isolated, one very crowded) are shown as open symbols with rightward arrows at the left boundary of panel (b), indicating unconstrained F275W−F336W colors; they are excluded from panel (a) where F336W enters both axes. The three crowding categories overlap substantially in CMD space, although a small number of very crowded targets extend to fainter magnitudes and/or redder colors. The remaining filter combinations (F475W−F814W, F814W−F110W, F110W−F160W) are presented in the online supplementary material.

A small subset of the very crowded sample (∼3\sim3 stars) appears at fainter F336W magnitudes and/or redder F336W–F475W colors. The redder colors are consistent with additional flux from cooler neighboring stars, which contribute relatively more at F475W than at F275W, while the fainter F336W magnitudes may reflect intrinsically fainter counterparts rather than crowding. Overall, the three crowding categories overlap substantially in CMD space. CMD position alone therefore provides limited leverage for identifying contaminated systems, compared with the spatial crowding metric in Figure 3.

For each target we also generated a local CMD comprising all PHATTER sources within 5′′5^{\prime\prime} (∼20\sim20 pc) in five filter combinations spanning the NUV through NIR (Figure 5).

Local color–magnitude diagrams and HST/PHATTER color composites for four representative Wolf-Rayet stars

Figure 5. Local CMDs (F475W vs. F336W−F475W) for the four representative WR stars shown in Figure 6. Each panel displays all PHATTER sources within 5′′5^{\prime\prime} (∼20\sim20 pc) of the WR position. Point color indicates local point-source density from a Gaussian kernel-density estimate, mapped from low (blue) through intermediate (yellow) to high (red). The identified WR counterpart (brightest F275W source within 1′′1^{\prime\prime}) is marked with the corresponding category symbol: green circle (isolated), orange square (moderately crowded), red diamond (very crowded), as defined in § 3.3. Insets: 5′′×5′′5^{\prime\prime} \times5^{\prime\prime} HST/PHATTER color composites (blue: F336W; green: F475W; red: F814W) rendered with a Lupton stretch that preserves relative surface brightness, giving a direct visual sense of the flux contributed by neighboring sources. The cyan circle marks the 0′′.750^{\prime\prime}.75-radius Hectospec fiber aperture. Panels: (a) isolated, (b) moderately crowded, (c, d) very crowded.

Ambiguous Cases

Our crowding analysis identified ∼10\sim10 targets for which the WR counterpart cannot be uniquely determined from broad-band imaging. These cases arise when two or more comparably bright (Δm12≲0.5\Delta m_{12} \lesssim0.5 mag) blue sources lie within the synthetic fiber aperture, each individually consistent with WR-type luminosities. In such configurations the fractional flux contributions are nearly degenerate and the ground-based centroid lies between competing candidates rather than on any individual source.

These systems fall into two categories:

  1. Close pairs: Two luminous blue stars separated by 0. ⁣ ⁣′′20.\!\!^{\prime\prime}2–0. ⁣ ⁣′′50.\!\!^{\prime\prime}5 (∼1\sim1–22 pc), each consistent with WR absolute magnitudes. In some cases both may be genuine WR stars—compact WR+WR or WR+O systems are known in the Milky Way and LMC [10, 48].

  1. Dense cluster cores: Targets embedded in the unresolved cores of compact young clusters or giant H II regions (e.g., NGC 604), where three or more comparable-brightness sources contribute to the fiber flux and even HST imaging cannot isolate the spectroscopic target.

Of the ∼10\sim10 ambiguous targets, roughly 4–6 fall into the close-pair category and the remainder are embedded in Note. Coordinates are J2000. Heliocentric radial velocities (vheliov_{\mathrm{helio}}) are derived from Keck/DEIMOS spectra. V-band magnitudes are from the catalog of Massey et al. (2006), updated in Neugent & Massey [35, 36].

dense cluster cores. These targets are flagged in Table 1 and illustrated in the postage-stamp figure (Figure 6). They are high-priority targets for AO-fed IFU spectroscopy; Keck/OSIRIS (LGS/NGS-AO) can spatially deblend these systems at ∼0.05′′\sim0.05''–0.1′′0.1'' resolution, substantially finer than the seeing-limited ∼0.3′′\sim0.3''–0.7′′0.7'' achievable with non-AO IFUs.

Four rows of HST/F336W postage stamps and PHATTER source maps for isolated, moderately crowded, ambiguous close-pair, and ambiguous cluster-core Wolf-Rayet stars

Figure 6. Representative HST/F336W postage stamps (7′′×7′′7^{\prime\prime} \times7^{\prime\prime}) and PHATTER source maps for four WR stars illustrating the three crowding categories defined in § 3.3. We display F336W imaging because it provides deeper, higher signal-to-noise photometry than F275W while preserving the ultraviolet contrast between hot WR stars and their cooler neighbors. Each row shows one WR star: the left panel displays the F336W image with the 0.75′′0.75^{\prime\prime}-radius synthetic fiber aperture (colored circle), the published ground-based catalog position (×), and the brightest PHATTER counterpart (+); the right panel shows all PHATTER sources within 3.5′′3.5^{\prime\prime}, with marker area proportional to F336W flux. Row 1 (green): An isolated WR star where a single bright source dominates the fiber flux. Row 2 (orange): A moderately crowded case where a second source contributes measurable flux within the fiber. Row 3 (purple): A very crowded ambiguous close pair where two comparably bright blue sources lie within the fiber; the WR spectrum cannot be assigned to either star from photometry alone. Row 4 (red): A very crowded case embedded in a dense compact association (ambiguous cluster core; § 3.5).

Serendipitous Keck/DEIMOS Spectroscopy of Five Wolf-Rayet Stars

During our systematic visual inspection of ≈7000\approx7000 TREX spectra (§ 2.3), we identified five previously cataloged WR stars that had been serendipitously placed on DEIMOS slitmasks. These objects—all classified as WN-type stars in the Neugent & Massey [35, 36] catalog—were not deliberately targeted as WR stars but were selected by the TREX color-magnitude criteria as luminous blue point sources in the PHATTER photometry. Their WN nature was immediately apparent from the strong, broad emission features in the extracted one-dimensional spectra, characteristic of dense, fast stellar winds ([10, 45]). The five targets and their basic properties are listed in Table 2.

Object Name (J2000)RA (deg)Dec (deg)V (mag)vheliov_{\mathrm{helio}} (km s−1^{-1})Spectral Type
J013327.26+303909.123.3634+30.652517.954−124.0Ofpe/WN9
J013332.97+304136.123.3873+30.693318.145...WNL
J013334.04+304117.223.3917+30.688119.779−37.2WN
J013353.60+303851.623.4732+30.647718.091−168.4Ofpe/WN9
J013418.74+303411.823.5780+30.569919.578−102.7Ofpe/WN

Table 2. Serendipitously Observed Wolf-Rayet Stars

Because the TREX target selection preferentially chose relatively isolated sources based on HST imaging, and because the Keck observations were obtained in sub-arcsecond seeing (≲0′′.8\lesssim0^{\prime\prime}.8), these spectra are substantially less contaminated by neighboring OB stars than the original Hectospec fiber spectra from which the WN classifications were derived. This makes them valuable modern benchmarks for the M33 WN population: while the broad WN classifications agree with those of Neugent & Massey [35, 36], the reduced continuum contamination enables more reliable measurements of individual spectral diagnostics, including emission-line

**Note—**EWs and FWHM are from Gaussian fits to continuum-normalized Keck/DEIMOS spectra. FWHM is converted to velocity via vFWHM=c Δλ/λ0v_{\mathrm{FWHM}} = c\,\Delta\lambda/\lambda_0. The WN and WNL objects lack clean coverage at He II λ4686 in the extracted spectra; He II λ5411 is used instead. Heliocentric velocities are from the cross-correlation pipeline (Table 2).

equivalent widths, line widths, and blue-edge velocities from P Cygni profiles (§ 4.3).

Spectral Classification Confirmation

Figure 7 presents the continuum-normalized Keck/DEIMOS spectra for the five WN-type stars in our sample, arranged in a spectral ladder spanning λλ4500–9500 Å. The sample illustrates the clear spectral progression of the WN sequence. The earlier, hotter WN and WNL subtypes (bottom traces) are dominated by strong, broad He II λ5411 emission, indicative of fast, highly ionized stellar winds. Conversely, the two objects at the cool end of the sequence—J013327.26+303909.1 and J013353.60+303851.6 (top traces)—display the relatively narrow emission lines and He I-dominated spectra characteristic of the Ofpe/WN9 transitional class. In these cooler winds, the He II features weaken significantly or vanish entirely, while He I lines (e.g., λ5876) develop distinct P-Cygni absorption troughs. These morphologies are consistent with their published catalog subtypes. We note that none of the five spectra show detectable C IV λ5808 emission, confirming the absence of WC characteristics and ruling out WN/WC transitional status.

Three of the five WR stars in our sample were reobserved with Keck/DEIMOS during follow-up programs in 2022 and 2023, providing a temporal baseline of 4–5 years relative to the original 2018 TREX observations. The follow-up observations used the 950G grating (λΔ4800–10000 Å), which does not cover the primary blue classification diagnostics (He II λ4686, N III λ4640), so the multi-epoch comparison is limited to He I λλ5876, 7065 and He II λ5411. The equivalent widths of these features for the two Ofpe/WN9 stars are consistent within ≲ 0.5 Å between the 2018B and 2022B/2023B epochs. We assessed radial-velocity stability by cross-correlating the He I λ7065 profile between same-grating (950G) epochs to avoid wavelength zero-point offsets between grating configurations. For J013353.60+303851.6 (Ofpe/WN9), observed in both 2022B and 2023B, no measurable velocity shift is detected over the one-year baseline, consistent with the instrumental precision of ∼30 km s−1^{-1} at R≈2000R \approx2000. Sparse multi-epoch sampling does not rule out long-period or low-amplitude binarity.

Nebular Hα Decontamination

Each DEIMOS slit extends several arcseconds along the spatial axis, enabling separation of the spatially extended narrow nebular Hα emission (FWHM ∼ 30–50 km s−1^{-1}) from the spatially unresolved broad stellar-wind component (FWHM ≳ 1000 km s−1^{-1}). The standard spec2d sky-subtraction procedure ([9]) models the extended emission from off-source rows

Rest-frame Keck/DEIMOS spectra of five WN-type Wolf-Rayet stars

Figure 7. Rest-frame Keck/DEIMOS spectra of the five WN-type Wolf-Rayet stars serendipitously observed during the TREX survey of M33, obtained with the 600ZD grating (R≈2000R \approx2000; λλ4500\lambda\lambda4500–95009500 Å). Spectra are continuum-normalized, offset vertically for clarity, and ordered by spectral subtype from earliest, hottest winds (bottom) to latest, coolest winds (top). Stellar wind features are marked with dotted lines and nebular emission features with dashed lines. The He II λ5411\lambda5411 broad emission lines—one of the primary early WN diagnostics—are prominent in the WN and WNL objects but weaken significantly or vanish in the late-type Ofpe/WN9 stars. Conversely, the two Ofpe/WN9 stars (top) exhibit narrower emission profiles and significant He I features (notably the P Cygni profile at He I λ5876\lambda5876) relative to the earlier WN subtype (bottom), consistent with their cooler, slower, and denser winds. The progression from broad, high-ionization features (WN) to narrow, low-ionization profiles (Ofpe/WN9) confirms the published WN classifications and, through reduced fiber contamination, enables quantitative wind diagnostics inaccessible from the original Hectospec data (§ 4.3, § 4.2).

and removes it from the extracted one-dimensional spectrum. Figure 8 illustrates this for J013332.97+304136.1 (WNL), the most severely contaminated target, where the narrow nebular component contributes ≈80%\approx80\% of the total HαH\alpha flux at the stellar position. Across the four slit-observed targets with adequate HαH\alpha coverage, nebular fractions range from a few percent to ∼80%\sim80\%, reflecting differences in local H II-region surface brightness. At sufficient spectral resolution (R≳1000R \gtrsim1000), the narrow nebular component can generally be distinguished spectrally; spatially resolved spectroscopy provides an additional, independent means of separation.

Nebular Hα contamination in the DEIMOS slit spectrum of J013332.97+304136.1 (WNL)

Figure 8. Nebular Hα contamination in the DEIMOS slit spectrum of J013332.97+304136.1 (WNL). (a) Two-dimensional slit image centered on Hα; the color scale maps flux from low (dark) to high (bright). The compact horizontal trace is the stellar continuum; the vertical features extending along the slit are the [N II] λ6548, 6583 and Hα nebular emission lines (white dotted lines mark their rest wavelengths). The cyan arrow indicates the WR stellar trace. (b) Spatial profiles at Hα (±12 Å; red) and in adjacent line-free continuum bands (6350–6450 Å and 6650–6750 Å; blue), both normalized to the continuum peak. The shaded region marks the excess of the Hα profile over the stellar PSF. (c) One-dimensional extracted spectrum. The dashed horizontal line marks the continuum level; red dotted vertical lines and labels identify the [N II] and Hα rest wavelengths. The narrow spike at the Hα rest wavelength is residual nebular emission that spatially overlaps the stellar PSF and survives local background subtraction. This target shows the largest nebular Hα fraction among the four slit-observed WR stars (§ 4.2).

Emission Line Properties

We measured equivalent widths, line widths, and wind velocities from the continuum-normalized Keck/DEIMOS spectra; results are summarized in Table 3.

ObjectTypeLineEW (Å)FWHM (km s−1^{-1})vblueav_{\mathrm{blue}}^{a} (km s−1^{-1})vheliov_{\mathrm{helio}} (km s−1^{-1})
J013334.04+304117.2WNHe II λ5411+8.31025 ± 38⋯−37
J013332.97+304136.1WNLHe II λ5411+9.2682 ± 141504b^{b}⋯
J013418.74+303411.8Ofpe/WNHe II λ4686+35.6562 ± 11755b^{b}−103
J013327.26+303909.1Ofpe/WN9He II λ4686−3.1c^{c}389 ± 67702b^{b}−124
He I λ5876+6.9432 ± 24⋯
J013353.60+303851.6Ofpe/WN9He II λ4686−3.6c^{c}⋯⋯−168
He I λ5876+2.8498 ± 27⋯

Table 3. Spectroscopic Properties of the Five TREX Wolf-Rayet Stars

We fitted single Gaussian profiles to the strongest He II feature available for each star: He II λ5411\lambda5411 for the WN and WNL objects, and He II λ4686\lambda4686 for the Ofpe/WN and Ofpe/WN9 subtypes. The measured vFWHMv_{\mathrm{FWHM}} values range from 329–1025 km s−1^{-1}, with the Ofpe/WN9 objects exhibiting systematically narrower profiles (vFWHM≈390v_{\mathrm{FWHM}} \approx390–560 km s−1^{-1}) than the WNL (682 km s−1^{-1}) and WN (1025 km s−1^{-1}) subtypes, consistent with the known progression from slow Ofpe/WN9 outflows to fast WNE winds (Crowther [10]). The two Ofpe/WN9 stars show He II λ4686\lambda4686 in net absorption (EW ≈−3\approx-3 Å), confirming their position at the O/WN boundary (Mathys [32]).

For lines exhibiting P Cygni profiles, we measured the blue-edge velocity vbluev_{\mathrm{blue}} from the blue edge of the absorption trough, defined as the wavelength at which the profile rises back to within 1σ1\sigma of the local continuum. P Cygni profiles were identified in He I λ5876\lambda5876 for three of the five targets, yielding vblue=702v_{\mathrm{blue}} = 702–1504 km s−1^{-1} (Table 3). Because He I λ5876\lambda5876 is a subordinate line and the absorption troughs are unsaturated, these blue-edge velocities do not necessarily correspond to the true v∞v_{\infty}; we therefore report them as He I blue-edge velocities throughout. For hotter WN subtypes, where He I is a minority ion, broad He II emission wings can mimic P Cygni morphology, and the measured vbluev_{\mathrm{blue}} values should be treated as upper limits.

Discussion

Our HST/PHATTER analysis quantifies, for the first time, the photometric contamination affecting individual M33 WR stars and provides Keck/DEIMOS slit spectra free from the fiber-blending and nebular contamination that has characterized previous ground-based work. In this section we discuss the implications of these results for spectral classification (§ 5.1), wind diagnostics (§ 5.2), and the broader use of M33 as a calibration anchor for population synthesis (§ ??), and we outline the catalog’s practical utility (§ 5.3).

Impact of Crowding on Spectral Classification

We find that 33 of 123 WR stars in the PHATTER footprint (∼27%) are classified as moderately or very crowded by our two-parameter metric (§ 3.3), with FWRF_{\mathrm{WR}} values typically below ∼0.85 and ∼0.50 for the two sub-samples, respectively. The sensitivity of the WN classification system to this dilution depends on which diagnostic is used.

Line-ratio classifications are largely preserved. The WN subtype sequence within the WNE and WNL classes is defined by ratios such as N III λ4640/He II λ4686 and He I λ5876/He II λ5411 ([50, 10]). Under continuum dilution, both numerator and denominator scale equally with FWRF_{\mathrm{WR}}, so line ratios are invariant.

Absolute EW thresholds, however, are directly affected. The boundary between WNL and the Ofpe/WN9 transitional class sits at EW(He II λ4686) ≈ 10–15 Å ([10, 32]). A WNL star with an intrinsic EW of 30 Å, typical of the class ([22]), would appear at 12 Å when observed through a fiber with FWR=0.4F_{\mathrm{WR}} = 0.4, marginally crossing into the Ofpe/WN9 regime. Our moderately and very crowded samples are therefore the most susceptible to systematic WNL→Ofpe/WN9 misclassification. A related effect was reported by Bibby & Crowther [4] for WR stars in starburst galaxies observed through ground-based seeing.

Our DEIMOS–Hectospec EW comparison (Figure 9) provides a direct empirical test. For the He II and He I λ6678 diagnostics the Hectospec EW is systematically smaller than the DEIMOS value, in the sense predicted by continuum dilution. Excluding the two cases discussed below, the measured ratios span H/D = 0.55–0.90, comparable in magnitude to the FWR475F_{\mathrm{WR}}^{475} values of the same targets. The agreement is not exact. J013332.97+304136.1 (WNL) shows H/D = 0.90±0.030.90 \pm0.03 at He II λ5411 against FWR475=0.75F_{\mathrm{WR}}^{475} = 0.75, while J013334.04+304117.2 (WN) shows H/D = 0.55±0.030.55 \pm0.03 against FWR475=0.80F_{\mathrm{WR}}^{475} = 0.80. The two deviations are of opposite sign and comparable size. With only four stars spectra available for direct comparison, we can confirm the direction and approximate magnitude of the dilution but cannot test the quantitative prediction H/D = FWRF_{\mathrm{WR}}. Individual FWRF_{\mathrm{WR}} values in Table 1 should therefore be treated as estimates of the dilution, not as bounds on it. A larger slit-spectroscopic sample is needed to determine whether the residual scatter reflects measurement uncertainty, the adopted seeing model, or low-level intrinsic variability.

Four-row, three-column comparison of continuum-normalized Keck/DEIMOS and MMT/Hectospec spectra for four Wolf–Rayet stars

Figure 9. Continuum-normalized Keck/DEIMOS (blue; slit, this work) and MMT/Hectospec (red; 1. ⁣′51.\!^{\prime}5 fiber, Massey et al.) spectra for the four WR stars, zoomed on three wind diagnostic regions. Each row corresponds to one target (labeled at left with spectral subtype); columns show He II λ5411\lambda5411, He I λ5876\lambda5876, and He I λ6678\lambda6678. Both spectra are continuum-normalized using linear fits to the grey-shaded sideband windows; the yellow-shaded region marks the EW integration window. Measured equivalent widths (EWD_{\mathrm{D}}, EWH_{\mathrm{H}}) and the Hectospec-to-DEIMOS ratio H/D are annotated in each panel. Panels marked “Hectospec: no coverage” indicate that the Hectospec wavelength solution does not extend to He I λ6678\lambda6678 for those observations. From top to bottom: J013334.04+304117.2 (WN; Very Crowded, FWR475=0.80F_{\mathrm{WR}}^{475}=0.80), J013332.97+304136.1 (WNL; Moderately Crowded, FWR475=0.75F_{\mathrm{WR}}^{475}=0.75), J013418.74+303411.8 (Ofpe/WN; Isolated, FWR475=0.97F_{\mathrm{WR}}^{475}=0.97), and J013327.26+303909.1 (Ofpe/WN9; Moderately Crowded, FWR475=0.83F_{\mathrm{WR}}^{475}=0.83).

J013418.74+303411.8 (Ofpe/WN) has the highest FWR475=0.97F_{\mathrm{WR}}^{475}=0.97 in the comparison sample, yet shows H/D = 0.40–0.53 across the three measured lines (Figure 9).

Continuum dilution cannot produce this. The Hectospec and DEIMOS epochs are separated by roughly a decade, and Ofpe/WN stars are known to vary on this timescale. The M33 catalog includes at least one object with an Ofpe/WN9 spectrum that is a confirmed LBV (Var 2). Intrinsic emission-line variability is the most likely explanation, though the low Hectospec S/N for this faint target (V=19.6V = 19.6) may also contribute. Second, the He I λ5876\lambda5876 measurement for J013334.04+304117.2 gives H/D = 5.26, with the Hectospec EW exceeding the DEIMOS value. The DEIMOS EW at this line is small (1.5 ±0.3\pm0.3 Å), so the ratio is poorly constrained, and the Hectospec profile shows a narrow component absent from the spatially resolved DEIMOS slit extraction. Because this star has the lowest nebular fraction in our sample (§ 4.2), we attribute the excess to a blended narrow feature in the fiber rather than to nebular recombination or the WR wind.

The dilution bias has a further consequence for the inferred surface hydrogen mass fraction XHX_{\mathrm{H}}, which discriminates between hydrogen-burning WNh stars (Minit≳80M_{\mathrm{init}} \gtrsim80–100 M⊙100\,M_{\odot}) and classical, hydrogen-depleted cWN stars ([18, 48, 45]). Continuum dilution weakens the broad He II λ4686\lambda4686 emission while preserving, or enhancing, hydrogen features contributed by OB neighbors. This artificially inflates XHX_{\mathrm{H}} and can mask the true evolutionary state of the star.

Our crowding classification enables a first-order correction. For the isolated sample (N=90N = 90), the classifications of Neugent & Massey [35, 36] can be considered reliable. For the moderate sample (N=13N = 13), published subtypes should be treated as lower limits on the true emission-line strength, with reclassification to earlier subtypes plausible after dilution correction. For the very crowded sample (N=20N = 20), spectral types are unreliable and spatially resolved follow-up is required for definitive classification.

Implications for Wind Diagnostics and Mass-Loss Rates

For emission-line-based M˙\dot{M} estimates, crowding introduces a systematic underestimate. Equivalent widths are diluted by FWRF_{\mathrm{WR}}; combined with WR-only continuum fluxes from HST photometry, this yields line luminosities underestimated by the same factor. Because recombination-line luminosities scale as M˙2\dot{M}^{2} in the optically thin regime ([10, 21]), an FWRF_{\mathrm{WR}} underestimate of LlineL_{\mathrm{line}} translates to an FWR1/2F_{\mathrm{WR}}^{1/2} underestimate of M˙\dot{M} when the bolometric luminosity and other stellar parameters are held fixed or independently constrained (e.g., from the deblended HST photometry provided here). If the bolometric luminosity is itself contaminated by neighbor flux, the corresponding bias in R∗R_{\ast} can partially compensate the EW effect through the transformed-radius scaling, and the net M˙\dot{M} bias may therefore be smaller. For our very crowded sample (FWR∼0.3F_{\mathrm{WR}} \sim0.3), the FWR1/2F_{\mathrm{WR}}^{1/2} estimate corresponds to a ∼45%\sim45\% bias in the fixed-luminosity limit. The F475W- and F814W-band FWRF_{\mathrm{WR}} values tabulated in the catalog Table 1 provide the appropriate correction factors for optical diagnostic lines (§ 3.2). Critically, because Hectospec is not flux-calibrated (Neugent & Massey [35]), the absolute line fluxes that would in principle survive continuum dilution were never available for the foundational M33 classifications.

Nebular Hα\alpha contamination introduces a second bias in fiber-based M˙\dot{M} estimates (§ 4.2). In practice, WR atmosphere models generally account for nebular contamination by excluding narrow nebular features from the fit, although this becomes more difficult when nebular and wind components overlap. The transformed radius Rt∝R∗(v∞/M˙)2/3R_{\mathrm{t}} \propto R_{\ast}(v_{\infty}/\dot{M})^{2/3} ([44, 21]) depends on both the stellar luminosity (and hence R∗R_{\ast}) and M˙\dot{M}. The six-band HST photometry provided here enables WR-only luminosity estimates and, together with the FWRF_{\mathrm{WR}} correction factors, reduces continuum-dilution systematics in future atmospheric fitting.

Our Keck/DEIMOS spectra provide empirical He I blue-edge velocity measurements (vblue≈700v_{\mathrm{blue}} \approx700–1500 km s−11500\,\mathrm{km\,s^{-1}}; Table 3). These measurements are less affected by continuum dilution than EW-based diagnostics, although they should not be interpreted as terminal wind velocities (§ 4.3). They provide useful constraints for future detailed atmosphere modeling with PoWR or CMFGEN.

Catalog Utility and Future Prospects

The clean sample of 90 WR stars with robust six-band HST photometry constitutes a high-fidelity training set for machine-learning classifiers identifying rare stellar types in large photometric surveys, while our flags for crowded and ambiguous sources can filter contaminated targets that would otherwise degrade automated classifications. Observational follow-up programs can preferentially target our isolated sample for clean wind diagnostics with seeing-limited IFUs or JWST/NIRSpec (where the higher spatial resolution mitigates nebular contamination relative to fiber spectroscopy), or our ambiguous and very crowded subsamples for definitive deblending via AO-fed IFU spectroscopy with Keck/OSIRIS (LGS/NGS-AO).

Finally, PHATTER covers only ∼50%\sim50\% of the M33 WR population. An extension to larger galactocentric radii would bring the remaining 89 outer-disk WR stars within reach of this analysis, where M33’s metallicity falls to ∼0.3 Z⊙\sim0.3\,Z_{\odot} [5]—precisely the regime where theoretical mass-loss predictions diverge most sharply [53].

Summary

We have presented a systematic crowding analysis of the M33 WR population using HST/PHATTER imaging, combined with new Keck/DEIMOS slit spectroscopy of five serendipitously observed WR stars. Our main results are:

  1. We developed a two-parameter crowding diagnostic (dd, Δm12\Delta m_{12}; § 3.3) and a complementary synthetic-fiber model (§ 3.2) that together categorize and quantify the fractional WR continuum contribution FWRF_{\rm WR} to the 1.′′51.''5 Hectospec fiber used for the foundational WR census. Applied to the 123 WR stars within the PHATTER footprint, the classification yields 90 isolated, 13 moderately crowded (FWRF_{\rm WR} typically ∼0.4\sim0.4–0.90.9), and 20 very crowded (FWRF_{\rm WR} typically ≲0.5\lesssim0.5) systems, where the classification is defined by the model-independent (dd, Δm12\Delta m_{12}) metric and FWRF_{\rm WR} is the corresponding fiber-model estimate (§ 3.3). Over one-quarter of the M33 WR population within PHATTER suffers significant continuum dilution from unresolved OB neighbors.

  1. We provide updated HST coordinates and six-band photometry (F275W, F336W, F475W, F814W, F110W, F160W) for 123 M33 WR stars, with the WR counterpart identified as the brightest F275W source within 1′′1'' of the catalog position (§ 3.1). Residual astrometric offsets between the corrected ground-based positions and the HST counterparts scale with crowding class (isolated: median 0.′′130.''13; very crowded: median 0.′′260.''26, maximum 0.′′880.''88), providing a geometry-based check consistent with the crowding classifications (Figure 2). The catalog is publicly available in machine-readable form (Table 1). The isolated sample occupies the expected luminous, blue locus on global CMDs, consistent with the crowding classification. A subset of very crowded stars is displaced toward redder colors, the photometric signature of unresolved OB continuum dilution in the optical bands.

  1. We identify ∼10\sim10 ambiguous cases in which two or more comparably bright blue sources lie within the fiber aperture and the WR counterpart cannot be uniquely assigned from broad-band imaging (§ 3.5). These arise primarily in the cores of NGC 604 and NGC 595. Because sub-arcsecond spatial resolution is essential for deblending these systems, they are high-priority targets for laser- or natural-guide-star AO-fed IFU spectroscopy; Keck/OSIRIS (LGS/NGS-AO) is particularly well-suited given M33’s northern declination (δ≈+30∘\delta\approx+30^\circ), which makes it a challenging target for VLT/MUSE.

  1. We present new Keck/DEIMOS spectra (λλ 4500\lambda\lambda\,4500–95009500 Å, R≈2000R \approx2000) for five WN-type stars serendipitously observed during the TREX survey (§ 4.1). Direct EW comparison with archival MMT/Hectospec spectra for four of these targets (Figure 9) confirms the predicted dilution: EWHecto/EWDEIMOS=0.55\mathrm{EW}_{\rm Hecto}/\mathrm{EW}_{\rm DEIMOS}=0.55–0.900.90, with the Hectospec values uniformly smaller. The sense and approximate magnitude of the effect match the FWRF_{\rm WR} prediction, but four stars are too few to test the relation EWH/EWD=FWR\mathrm{EW}_{\rm H}/\mathrm{EW}_{\rm D}=F_{\rm WR} quantitatively.

  1. We measure He II line widths (vFWHM≈330v_{\rm FWHM} \approx330–1025 km s−11025\ {\rm km\,s^{-1}}; Table 3) and P Cygni blue-edge velocities of He I absorption (vblue≈700v_{\rm blue} \approx700–1500 km s−11500\ {\rm km\,s^{-1}}), detected in three of five targets from the DEIMOS spectra (§ 4.3). The Ofpe/WN9 subtypes show systematically narrower profiles than the WN/WNL subtypes, consistent with their cooler, denser winds. These measurements provide modern wind benchmarks for the M33 WN population at its intermediate metallicity (Z≈0.5Z \approx0.5–1.0 Z⊙1.0\,Z_{\odot}), and are robust against continuum dilution since they are referenced to the local continuum rather than to integrated line fluxes.

  1. Multi-epoch Keck/DEIMOS spectra spanning 4–5 years for three of the five targets reveal no statistically significant EW or radial-velocity variability (ΔEW≲0.5\Delta\mathrm{EW} \lesssim0.5 Å, ∣Δv∣≲30 km s−1|\Delta v| \lesssim30\ {\rm km\,s^{-1}}; § 4.1). This temporal stability argues against short-period binary modulation (P≲10P \lesssim10 yr) and rules out the eruptive variability characteristic of LBV impostors, reinforcing the classification of these objects as classical, stable WR stars.

  1. For the four WR stars with clean DEIMOS slit extractions, the two-dimensional spectral image reveals spatially extended [N II] and Hα\alpha emission from the surrounding H II region, with nebular fractions at the stellar position ranging from $\lesssim 5%

to ∼80%\sim80\% depending on the local nebular environment (§ 4.2). In MMT/Hectospec fiber spectra, the nebular component is blended with the stellar wind emission, systematically inflating the apparent HαH\alpha luminosity and biasing HαH\alpha-based mass-loss rate determinations. This effect is irremovable in fiber spectra but can be spatially separated with slit or IFU spectroscopy.

The crowding-corrected catalog and the new slit spectra together provide a foundational reference for reevaluating M33’s role as a calibration anchor for population synthesis models, for future spatially resolved follow-up of the ambiguous and very crowded subsamples, and for the construction of high-fidelity photometric training sets for machine-learning WR classifiers.

Data Availability

The full machine-readable catalog (Table 1), containing HST/PHATTER positions, six-band photometry (F275W–F160W, VEGAMAG), crowding classifications, FWRF_{\mathrm{WR}} values, and spectral cross-identifications for all 123 WR stars within the PHATTER footprint, is available via the journal’s online supplement and at M33-WR-Crowding, which also contains the Keck/DEIMOS one-dimensional extracted spectra (FITS format) for the five WR stars presented in § 4. The underlying PHATTER source catalogs are publicly available on MAST: doi:10.17909/t9-ksyp-na40 (Williams et al. [55]).6

HST/F275W, F336W, F475W, and F160W postage stamps and local color-magnitude diagrams for all 123 WR stars are provided as figure sets in the online journal (Figure 6 and Figure 5).

Acknowledgments

We thank the anonymous referee for a constructive report that improved the clarity of this paper. We thank Philip Massey for helpful discussions, suggestions, and access to Hectospec spectroscopy data from the M33 WR census. We thank Matt Ashby and Andrea Dupree for a careful reading of the manuscript and helpful suggestions that improved the presentation. We thank Olivia Gaunt and Preksha Sethia for their contributions to the early stages of this project. This work is based on observations made with the NASA/ESA Hubble Space Telescope, obtained from the Mikulski Archive for Space Telescopes (MAST) at the Space Telescope Science Institute. Support for Program HST GO-14610 was provided by NASA through a grant from the Space Telescope Science Institute under NASA contract NAS5-26555.

The spectroscopic data presented herein were obtained at the W. M. Keck Observatory. The authors wish to recognize and acknowledge the very significant cultural role and reverence that the summit of Maunakea has always had within the indigenous Hawaiian community. We are most fortunate to have the opportunity to conduct observations from this mountain.

The TREX spectroscopic survey was supported by National Science Foundation grants AST-1909066 (KMG) and AST-1909759 (PG).

Software: astropy (Astropy Collaboration et al. [1, 2, 3]), numpy (Harris et al. [23]), matplotlib (Hunter [26]), scipy (Virtanen et al. [54]), spec2d/spec1d (Cooper et al. [9]; Newman et al. [37]).

Facilities: HST (WFC3/UVIS, ACS/WFC, WFC3/IR), Keck II (DEIMOS), MMT (Hectospec).

ORCID IDS

Rohit Raj https://orcid.org/0009-0000-0170-6302

Puragra Guhathakurta https://orcid.org/0000-0001-8867-4234

Robert E. Williams https://orcid.org/0000-0002-3742-8460

Benjamin F. Williams https://orcid.org/0000-0002-7502-0597

Julianne J. Dalcanton https://orcid.org/0000-0002-1264-2006

Meredith J. Durbin https://orcid.org/0000-0001-7531-9815

References

  1. [1]Astropy Collaboration, Robitaille, T. P., Tollerud, E. J., et al. 2013, A&A, 558, A33
  2. [2]Astropy Collaboration, Price-Whelan, A. M., Sipőcz, B. M., et al. 2018, AJ, 156, 123
  3. [3]Astropy Collaboration, Price-Whelan, A. M., Lim, P. L., et al. 2022, ApJ, 935, 167
  4. [4]Bibby, J. L., & Crowther, P. A. 2010, MNRAS, 405, 2737
  5. [5]Bresolin, F. 2011, ApJ, 730, 129. https://doi.org/10.1088/0004-637X/730/2/129
  6. [6]Cerviño, M., & Mas-Hesse, J. M. 1994, A&A, 284, 749
  7. [7]Chiosi, C., & Maeder, A. 1986, ARA&A, 24, 329
  8. [8]Conti, P. S. 1984, in IAU Symposium, Vol. 105, Observational Tests of the Stellar Evolution Theory, ed. A. Maeder & A. Renzini, 233DOI
  9. [9]Cooper, M. C., Newman, J. A., Davis, M., Finkbeiner, D. P., & Gerke, B. F. 2012, Astrophysics Source Code Library, ascl
  10. [10]Crowther, P. A. 2007, ARA&A, 45, 177
  11. [11]Cullinane, L. R., Gilbert, K. M., Guhathakurta, P., et al. 2023, ApJ, 958, 157
  12. [12]Drissen, L., Moffat, A. F. J., & Shara, M. M. 1993, AJ, 105, 1400
  13. [13]Eldridge, J. J., & Stanway, E. R. 2009, MNRAS, 400, 1019
  14. [14]Faber, S. M., Phillips, A. C., Kibrick, R. I., et al. 2003, in Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Series, Vol. 4841, Instrument Design and Performance for Optical/Infrared Ground-based Telescopes, ed. M. Iye & A. F. M. Moorwood, 1657–1669
  15. [15]Fabricant, D., Fata, R., Roll, J., et al. 2005, PASP, 117, 1411
  16. [16]Freedman, W. L., Wilson, C. D., & Madore, B. F. 1991, ApJ, 372, 455
  17. [17]Gilbert, K. M., Quirk, A. C. N., Guhathakurta, P., et al. 2022, ApJ, 924, 116
  18. [18]Gräfener, G., & Hamann, W. R. 2008, A&A, 482, 945
  19. [19]Hainich, R., Pasemann, D., Todt, H., et al. 2015, A&A, 581, A21
  20. [20]Hainich, R., Rühling, U., Todt, H., et al. 2014, A&A, 565, A27
  21. [21]Hamann, W.-R., Gräfener, G., & Liermann, A. 2006, A&A, 457, 1015
  22. [22]Hamann, W.-R., Gräfener, G., Liermann, A., et al. 2019, A&A, 625, A57
  23. [23]Harris, C. R., Millman, K. J., van der Walt, S. J., et al. 2020, Nature, 585, 357
  24. [24]Hillier, D. J., & Miller, D. L. 1998, ApJ, 496, 407
  25. [25]Hirschi, R., Meynet, G., & Maeder, A. 2005, A&A, 443, 581
  26. [26]Hunter, J. D. 2007, Computing in Science and Engineering, 9, 90
  27. [27]Lee, A. J., Rousseau-Nepton, L., Freedman, W. L., et al. 2022, ApJ, 933, 201. https://doi.org/10.3847/1538-4357/ac7321
  28. [28]Massey, P. 1985, PASP, 97, 5
  29. [29]Massey, P., & Johnson, O. 1998, ApJ, 505, 793
  30. [30]Massey, P., Neugent, K. F., Dorn-Wallenstein, T. Z., et al. 2021, ApJ, 922, 177
  31. [31]Massey, P., Neugent, K. F., & Smart, B. M. 2016, AJ, 152, 62. https://doi.org/10.3847/0004-6256/152/3/62
  32. [32]Mathys, G. 1988, A&AS, 76, 427
  33. [33]Meynet, G., & Maeder, A. 2005, A&A, 429, 581
  34. [34]Mighell, K. J. 2005, MNRAS, 361, 861 https://doi.org/10.1111/j.1365-2966.2005.09208.x
  35. [35]Neugent, K. F., & Massey, P. 2011, ApJ, 733, 123
  36. [36]Neugent, K. F., & Massey, P. 2014, ApJ, 789, 10
  37. [37]Newman, J. A., Cooper, M. C., Davis, M., et al. 2013, ApJS, 208, 5
  38. [39]Quirk, A. C. N., Guhathakurta, P., Gilbert, K. M., et al. 2022, AJ, 163, 166
  39. [40]Ramachandran, V., Hamann, W. R., Oskinova, L. M., et al. 2019, A&A, 625, A104
  40. [41]Regan, M. W., & Vogel, S. N. 1994, ApJ, 434, 536
  41. [42]Rosolowsky, E., & Simon, J. D. 2008, ApJ, 675, 1213
  42. [43]Sander, A. A. C., Hamann, W.-R., Todt, H., et al. 2019, A&A, 621, A92
  43. [44]Schmutz, W., Hamann, W.-R., & Wessolowski, U. 1989, A&A, 210, 236
  44. [45]Shenar, T. 2024, in IAU Symposium, Vol. 361, Massive Stars Near and Far, ed. J. S. Mackey, J. S. Vink, & N. St-Louis, 465–472
  45. [46]Shenar, T., Gilkis, A., Vink, J. S., Sana, H., & Sander, A. A. C. 2020, A&A, 634, A79
  46. [47]Shenar, T., Sablowski, D. P., Hainich, R., et al. 2019, A&A, 627, A151
  47. [48]Shenar, T., Sana, H., Marchant, P., et al. 2021, A&A, 650, A147
  48. [49]Simon, J. D., & Rosolowsky, E. 2008, in Astronomical Society of the Pacific Conference Series, Vol. 396, Formation and Evolution of Galaxy Disks, ed. J. G. Funes & E. M. Corsini, 23
  49. [50]Smith, L. F., Michael M., S., & Moffat, A. F. J. 1996, MNRAS, 281, 163. https://doi.org/10.1093/mnras/281.1.163
  50. [51]Smith, N. 2014, ARA&A, 52, 487
  51. [52]Tonry, J., & Davis, M. 1979, AJ, 84, 1511
  52. [53]Vink, J. S., de Koter, A., & Lamers, H. J. G. L. M. 2001, A&A, 369, 574
  53. [54]Virtanen, P., Gommers, R., Oliphant, T. E., et al. 2020, Nature Medicine, 17, 261
  54. [55]Williams, B. F., Durbin, M. J., Dalcanton, J. J., et al. 2021, ApJS, 253, 53
  55. [56]Woosley, S. E., & Heger, A. 2006, in American Institute of Physics Conference Series, Vol. 836, Gamma-Ray Bursts in the Swift Era, ed. S. S. Holt, N. Gehrels, & J. A. Nousek (AIP), 398–407
  56. [57]Zinnecker, H., & Yorke, H. W. 2007, ARA&A, 45, 481

Paper details

vlnc:W_57xgz64a
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents