Gravitational lensing reveals ionizing ultraviolet photons escaping from a distant galaxy

See allHide authors and affiliations

Science  08 Nov 2019:
Vol. 366, Issue 6466, pp. 738-741
DOI: 10.1126/science.aaw0978

Ionizing photons escape a lensed galaxy

Young, hot stars emit ultraviolet radiation, which can ionize a neutral gas. The first generation of stars converted most of the intergalactic gas in the Universe from neutral to ionized form during the epoch of reionization less than a billion years after the Big Bang. Rivera-Thorsen et al. took advantage of a gravitational lensing system to observe 12 images of the same star-forming region in a distant galaxy and determined the fraction of ultraviolet photons that escape into the intergalactic medium. Although this galaxy is younger than the epoch of reionization, the results provide clues about how ultraviolet photons escape their host galaxies and contribute to the reionization process.

Science, this issue p. 738


During the epoch of reionization, neutral gas in the early Universe was ionized by hard ultraviolet radiation emitted by young stars in the first galaxies. To do so, ionizing ultraviolet photons must escape from the host galaxy. We present Hubble Space Telescope observations of the gravitationally lensed post-reionization galaxy PSZ1-ARC G311.6602–18.4624 (nicknamed the “Sunburst Arc”), revealing bright, multiply imaged ionizing photon escape from a compact star-forming region through a narrow channel in an optically thick gas. The gravitational lensing magnification shows how ionizing photons escape this galaxy, contributing to the reionization of the Universe. The multiple sight lines to the source probe absorption by intergalactic neutral hydrogen on a scale of less than a few hundred parsecs.

Less than a billion years after the Big Bang, the Universe went through the epoch of reionization (EoR) (1), in which Lyman continuum radiation (LyC, ultraviolet light with wavelengths below 912 Å, which is capable of ionizing hydrogen) from the first galaxies ionized the previously neutral hydrogen in the intergalactic medium (IGM). To escape into the IGM, the ultraviolet photons must have avoided absorption by neutral hydrogen within the host galaxy. After the Universe became ionized and transparent to ionizing wavelengths, we expect that many galaxies would have continued to emit LyC photons. However, only a few dozen such galaxies have been found, either in the local Universe (25) or at intermediate redshifts (1 ≲ z ≲ 4) (610), leaving much of the radiation necessary to reionize the Universe unaccounted for.

Escape of LyC photons from galaxies is made possible by radiative or mechanical feedback from, e.g., turbulence or young, hot stars, which can ionize most of the surrounding gas or carve out channels through optically thick neutral gas (1115). These escape scenarios can be distinguished by the spectral shape of the Lyman α (Lyα) emission feature, an atomic emission line arising from the transition between the ground and first excited state in neutral hydrogen, at a wavelength of 1216 Å. Lyα scatters resonantly in the same neutral hydrogen that absorbs LyC, so the Lyα line shape is sensitive to the gas kinematics, geometry, and other properties (13). Escape through an optically thin medium results in a double-peaked Lyα profile with narrow separation between the peaks (3, 16, 17). That line shape is typically found in conjunction with confirmed LyC escape, except in a few ambiguous cases (3, 7).

In a clumpy, optically thick neutral medium, both Lyα and LyC may escape freely through empty lines of sight, with little to no interaction with neutral hydrogen. Lyα then appears as a bright, narrow peak centered at the wavelength of the transition. If the neutral medium contains enough passageways of low optical depth, then the majority of Lyα photons escape after scattering a small number of times, so the narrow central peak dominates the line shape (13). If only few narrow channels are present, then theoretical models predict that the probability for a given Lyα photon to escape in a few scattering events declines and the probability of trapping in resonant scattering in the denser neutral medium rises (13). This produces a characteristic, triple-peaked profile with a narrow, bright peak at the line center superimposed on the typical broader, double-peaked line of an optically thick medium (13, 18).

Previous work has observed such a triple-peaked Lyα profile in PSZ1-ARC G311.6602–18.4624, hereafter referred to by the nickname “Sunburst Arc” (13), a gravitationally lensed galaxy at redshift z = 2.37 (13). The Sunburst Arc is a single galaxy lensed into at least 12 images by a massive foreground cluster of galaxies at z = 0.44 (19). The galaxy is young, strongly star forming, and shows no sign of an active nucleus (fig. S1). The Lyα profile indicated probable strong LyC escape through a narrow channel oriented toward Earth (13). Such a LyC channel should appear as a multiply imaged, compact source coincident with some of the brightest regions seen in the extended, non-ionizing stellar continuum (19) when observed using telescopes with sufficient resolution at rest-frame ultraviolet wavelengths.

We observed the rest-frame LyC in the Sunburst Arc using Wide Field Camera 3 (WFC3) on the Hubble Space Telescope (HST) with the broadband filter F275W. The wavelength cutoff of this filter matches the lowest-energy limit of the LyC at the redshift of the lensed galaxy, with only 0.5% of the total filter throughput at wavelengths longer than 912 Å. We combined the F275W observations with previous observations using the HST Advanced Camera for Surveys (ACS) with the broadband F814W filter. At the redshift of the lensed galaxy, F814W is sensitive to the non-ionizing near-UV light emitted mostly by the same young, hot stars as LyC, but is not absorbed by neutral hydrogen.

Figure 1A shows the F814W image of the lens and arc system and Fig. 1, B to K, shows close-ups of regions with emission in the F275W filter. Each region is shown in both the ionizing and non-ionizing wavelengths, with the multiple images of the LyC emitting complex marked in both filters for comparison. Image 5 is contaminated by UV continuum from a foreground galaxy, which contributes ≲5% to its measured flux in F814W. Lens modeling and spectroscopy show that the 12 image-plane sources of ionizing LyC emission are all lensed images of the same bright region [fig. S2 and (20)], with signal-to-noise ratios ranging from 4 to 42 (Table 1). The physical diameter of the LyC emitting region has an upper limit of ~160 pc (20), much smaller than the galaxy as a whole and consistent with star-forming regions in local galaxies (21).

Fig. 1 Comparison of ionizing and non-ionizing morphology.

(A) Overview of the entire arc in the F814W filter with cutout locations marked. The surrounding panels (B to K) show paired F814W and F275W observations zoomed in on the regions with confirmed LyC detection. All panels are oriented with north up and east to the left. Data in panels (B) to (K) are stretched by a hyperbolic arcsine function to balance visibility of faint features in F814W with the noise in F275W panels. Panel (A), in which noise is negligible, is scaled by a square root function to better enhance faint features and colors are inverted. All F814W panels except (A) share the same cut levels. All F275W panels share the same cut levels, except (E) and (I), which are cut off at 60% of that level because of the fainter sources. The bright object near image 12 is a foreground star.

Table 1 Properties of regions with detected LyC escape.

The first column lists the lensed image numbers as designated in Fig. 1. The second and third columns show the measured apparent magnitude mABF275Wand signal-to-noise ratio (S/NF275W) of the LyC (F275W) observations. The fourth column shows the apparent magnitudes mABF814Win non-ionizing UV (F814W). The fifth column shows the apparent relative escape fractions (see main text) measured for each image. The corresponding apparent absolute escape fraction can be found by multiplying the relative escape fractions by a factor of 0.34 ± 0.04, which encodes the effect of internal dust in the galaxy (20). The sixth and final column shows the celestial right ascension (RA) and declination (Dec) in hour:minute:second, degree:arcminute:arcsecond, J2000 equinox. Apparent magnitudes have been corrected for foreground Milky Way dust reddening. Flux uncertainties in F814W are all <0.1%.

View this table:

Measured magnitudes in both filters are shown in Table 1, along with the computed apparent escape fraction. We computed ionizing escape fractions based on theoretical models of stellar populations (20), which were fitted to spectroscopic observations of the non-ionizing wavelengths emitted (13, 20) by the star-forming region. From these model spectra, we predicted the intrinsic flux ratios in the F275W and F814W filters and compared these with the observed ratios (20). We derived both the relative and absolute escape fractions, defined as the fraction of dust-attenuated (relative) and total (absolute) ionizing radiation that escapes the neutral gas in the galaxy.

The observed flux in F275W is the radiation surviving absorption both within the source galaxy and in the IGM. Consequently, the escape fraction that we derive from the flux ratios in the two filters is the combined effect of the internal and intergalactic neutral hydrogen (H, I). This quantity, found as the escape fraction from the galaxy (fesc) times the transmission coefficient of the IGM (TIGM), we refer to as the apparent escape fraction fesc*. The highest measured apparent escape fraction of 64 ± 9% (Fig. 1, J and K, image 12) forms a lower limit to the escape fraction, corresponding to the case of a completely transparent IGM. Conversely, the apparent escape fraction in image 12 provides a lower limit to the IGM transmission of TIGM ≳ 64%, as lower transmission coefficients would imply an escape fraction higher than 100%. To further constrain the line-of-sight escape fraction, we used the TIGM distribution along simulated lines of sight (22) and excluded the values that would lead to an escape fraction larger than 100%. From the remaining TIGM distribution, we have extracted the 16th, 50th, and 84th percentiles and computed the corresponding escape fractions for image 12 (20). Figure 2 shows the apparent relative (fesc,rel*) and absolute (fesc,abs*) ionizing escape fractions along our line of sight for each lensed image, with propagated measurement uncertainties. Absolute and relative escape fractions are shown for image 12 based on the IGM transmission distribution discussed above, the 16th and 84th percentiles, and the full range bracketed by an escape fraction of 100% and the far upper end of the TIGM distribution (20). We estimate a relative line-of-sight ionizing escape fraction of fesc,rel=9311+7%, with 46% as a robust lower limit (assuming a completely transparent IGM and a flux ratio 2σ below the measured value). Correcting for modeled dust absorption internal to the galaxy, we find the corresponding absolute line-of-sight escape fraction fesc,abs=324+2%. However, there is not a simple correspondence between the line-of-sight escape fraction often reported in the literature and the global ionizing escape fraction from a galaxy unless the galaxy is perfectly isotropic. The Lyα profile of the Sunburst Arc (13) indicates that its global escape fraction is much lower than the line-of-sight value that we report.

Fig. 2 LyC escape fractions.

For each gravitationally lensed image, we show the fraction of the dust-attenuated (relative; filled circles) and total (absolute; empty circles) LyC photons that reach the telescope (apparent escape fraction, fesc*). Colors correspond to those in Fig. 1. The black and white box-and-whiskers plots for image 12 show the median value (dot), 16th to 84th percentile (box), and full allowed range (whiskers) of the relative (filled) and absolute (outlined) escape fraction corrected for modeled IGM absorption. Values are listed in Table 1. Systematic uncertainties are not included but are discussed in the supplementary materials (20).

The computed escape fractions are subject to systematic uncertainties. The spectra used to construct the stellar populations are extracted in a larger aperture than the photometric measurements, which could affect the measured escape fractions (20). The F814W filter, which is used to photometrically calibrate the stellar population model and predict the intrinsic LyC flux, covers an adjacent but different wavelength range to the spectra used to model the stellar population. Therefore, we cannot directly test the accuracy of the model in the given range when comparing it with the photometric observations. Assumed stellar population models and dust attenuation laws introduce systematic uncertainties that can affect the inferred escape fractions, but not the differences between them (20).

Lensing models [figs. S2 and S3 and (20)] show that all ionizing sources in arcs 1, 3, and 4 (the counterarc) are lensed images of the same system. Arc 2 has a complicated lensing geometry that has not been modeled completely, but from the other arc segments, we find it likely that the ionizing detections in arc 2 are also images of the same system. This is supported by a comparison of rest-frame near-UV spectra of five locations in arcs 1 and 2: four that have detected LyC and one that does not. Comparison of the stellar triple-ionized carbon (C IV) 1550 Å and interstellar and circumgalactic triple-ionized silicon (Si IV) 1393,1402 Å features show that the four ionizing locations are indistinguishable, whereas the non-ionizing location is different [fig. S2 and (20)].

The lensing models indicate that the magnification factor in arc 1 is between 10 and 30 for each image. The central LyC source is unresolved in all images, which places an upper limit on the source size equal to the instrument point spread function of 0.9′′, corresponding to ~500 pc in the lens plane. For a magnification of 10 (in area), this corresponds to a physical diameter of ~160 pc in the source plane. With a magnification of 30, the maximum diameter would be ~90 pc, consistent with the sizes of star-forming regions in local galaxies (21) and at higher redshifts (21, 23).

Light from the multiple lens-plane images of the source galaxy traverses different paths between their point of origin and Earth. We interpret the variations in fesc* as being due to varying amounts of absorbing neutral hydrogen along these paths. The absorption most likely occurs at z ≲ 2, at which half of the LyC emission detected in the F275W image has been redshifted beyond the ionization limit of hydrogen (20). In Fig. 3, we show the transverse distance between lines of sight toward images 2 and 3 (one of the closest pairs), 1 and 6 (typical size of an arc segment), and 1 and 12 (on opposite sides of the arc) as a function of redshift, and mark the transverse distance between the lines of sight toward images 2 and 3 at z = 2. At the latter redshift, the physical transverse distance between the lines of sight to images 2 and 3 is d ≈ 330 pc.

Fig. 3 Transverse physical distances between lines of sight.

Transverse physical separation d of lines of sight with angular separations θ of 1, 10, and 55 arcseconds (corresponding to the distances between images 2 and 3, 1 and 6, and 6 and 12, respectively) in the lensing plane, shown as a function of redshift. The redshifts of the lensing plane (z = 0.44), the lensed galaxy (z = 2.37), and the minimum redshift of intervening H, I absorption (z ≲ 2.0) are marked, along with the maximum transverse distances at the point of H, I absorption.

The upper limit on the redshift of absorption depends on the combined extent of the ionizing source(s). For diameters of 160 or 30 pc, the absorption must occur at redshifts of z ≲ 2.27 or z ≲ 2.35, respectively. Even for a very compact cluster a few parsecs across, the absorption must still be ≳1 Mpc outside the galaxy. Possible absorbers include an undetected, interloping galaxy or circumgalactic or intergalactic systems of neutral hydrogen. It is also possible that the LyC originates in just one (or a few) very massive O-type or Wolf–Rayet stars (24), in which case the absorption could occur in the interstellar or circumgalactic medium of the source galaxy itself, and the transverse separation of sight lines could be <1 pc.

We considered differential magnification as a possible alternative explanation of the differences in apparent escape fraction. However, this effect would lead to a correlation between observed F814W magnitudes and fesc*. With a Pearson’s r = 0.2 and a significance of p = 0.53, we do not find such a correlation [fig. S5 and (20)].

Supplementary Materials

Materials and Methods

Supplementary Text

Figs. S1 to S6

References (2856)

Data S1

References and Notes

  1. Materials and methods are available as supplementary materials.
Acknowledgments: E.R.-T. thanks Stockholm University for their kind hospitality, M. Hayes and A. Adamo for informative and helpful conversations, and K. Vasei and coauthors for generously sharing their intermediate science products. Funding: E.R.-T. and H.D. acknowledge support from the Research Council of Norway. M.G. was supported by NASA through NASA Hubble Fellowship grant no. HST-HF2-51409 awarded by the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., for NASA under contract no. NAS5-26555. J.R. was funded as a NASA civil servant, with additional funding provided through NASA Hubble Space Telescope Guest Observer grants. Author contributions: E.R.-T. led the project and writing of the manuscript, with input and feedback from all coauthors, especially H.D. and M.G. E.R.-T. made the figures. H.D. wrote the HST proposals leading to the F275W and F814W observations, assisted by E.R.-T. M.K.F. reduced and combined the images in both filters. E.R.-T. performed photometry and computed escape fractions and transverse line-of-sight distances. J.C. performed stellar population synthesis based on spectroscopic observations made by J.R. and M.B and reduced by J.R. M.D.G. constructed the spatial model of flux in the MagE aperture. K.S. and G.M. produced the lens model. Competing interests: The authors declare no competing interests. Data and materials availability: Raw Hubble Space Telescope WFC3 F275W and ACS F814W observations with calibration data are available at the Mikulski Archive for Space Telescopes under proposal IDs 15418 (F275W) and 15101 (F814W). Raw Magellan/MagE observations and corresponding calibration data are available at Figshare (25). The lens model was generated with the publicly available code LENSTOOL (26); our model parameter file, along with a DS9 region file showing the critical curves of the model, is available in supplemental data S1. Scripts used to generate the stellar population models are available at Zenodo (27).

Stay Connected to Science

Navigate This Article