NICER Discovers Spectral Lines during Photospheric Radius Expansion Bursts from 4U 1820−30: Evidence for Burst-driven Winds

Thermonuclear X-ray bursts provide unique opportunities to study the extreme physics of neutron stars. These bright X-ray events result from unstable burning in an accreted layer of light elements on the neutron star's surface. The strongly heated surface layers briefly shine brightly in the X-ray band, and as the radiation comes directly from the stellar surface, it provides important clues regarding neutron star structure and the nuclear burning on them (see Strohmayer & Bildsten 2006; Galloway & Keek 2017, for reviews). The continuum emission during bursts is broadly consistent with blackbody emission from the heated surface layers. A major focus of recent efforts has been to employ the observed spectra to constrain the stellar mass and radius (Lewin et al. 1993). The basic idea is to measure an X-ray flux, F_x, and color temperature, T_col, and then use an appropriate atmosphere model to convert the color temperature to an effective temperature, T_eff. One can then, in principle, infer an assumed spherical emission area from the Stefan Boltzmann law, $L=4\pi {d}^{2}{F}_{x}=4\pi {R}^{2}\sigma {T}_{\mathrm{eff}}^{4}$, where d and R are the source distance and photospheric radius, respectively. For bursts that reach the Eddington luminosity, an additional constraint is available, as this quantity depends only on the stellar mass, radius, and the composition of the atmosphere. A full treatment must also account for the significant gravitational redshift from the neutron star surface, as well as any special relativistic effects associated with the potentially rapid spin of its surface. Substantial efforts have been made by a number of research groups over the past decade to carry out this program (see, for example, Özel 2006; Steiner et al. 2010; Güver et al. 2016; Nättilä et al. 2017, and references therein). While very promising, significant uncertainties remain in interpreting such observations. These can be illustrated with a few rhetorical questions: how accurate is the atmosphere model, in particular its composition? Are other emission components, such as, for example, reprocessed emission from an accretion disk, present in the observed spectrum? How accurately are the distance, flux, and color temperature known?

Another way to obtain constraints on the stellar mass and radius is to infer a surface gravitational redshift, z = (E₀ − E)/E, by observing a spectral feature; a line or edge at energy E that originates from the stellar surface, of a known, rest-frame transition with energy E₀. A constraint on the mass-to-radius ratio, the "compactness," M/R, then follows from the relation 1 + z = (1 − 2GM/c²R)^−1/2. While perhaps conceptually simpler than correctly interpreting observed continuum spectra, questions remain. Is the line identification secure? Is the emission site at the stellar surface, or some other altitude or location? Perhaps most fundamental of all is that it has been very difficult to reliably identify any surface spectral features in X-ray burst spectra.

There were some early claims of absorption line detections from the bursters 4U 1636−536, 4U 1608−522, and EXO 1747−214 based on data from Tenma and EXOSAT (Waki et al. 1984; Nakamura et al. 1988; Magnier et al. 1989), but subsequent observations with ASCA, RXTE, and BeppoSAX did not tend to confirm these initial reports. There has also been a recent report based on NuSTAR observations of GRS 1741.9−2853 of an absorption line at 5.46 ± 0.1 keV during an X-ray burst from this object (Barrière et al. 2015). However, the reported significance of the line was rather modest (1.7σ). Observations of two superbursts, from 4U 1820−30 and 4U 1636−536, with the RXTE Proportional Counter Array (PCA), clearly show emission features in the Fe K band that have been successfully modeled as burst thermal emission reprocessed (reflected) by the accretion disk (Strohmayer & Brown 2002; Keek et al. 2014a, 2014b). While these features provide an exciting view of the interaction of the burst flux with the accretion disk (Ballantyne & Strohmayer 2004; Fragile et al. 2018), it seems highly likely that they do not originate at the stellar surface.

Cottam et al. (2002) reported on a set of narrow absorption lines in the coadded, high-resolution spectra of 28 bursts from EXO 0748−676 observed with the XMM-Newton Reflection Grating Spectrometer. They argued that these features could be associated with transitions of Fe xxv, Fe xxvi, and O viii at a consistent surface redshift of 1 + z = 1.35. Unfortunately, subsequent observations did not confirm these lines, and the likely stellar spin frequency was determined to be 552 Hz (Galloway et al. 2010), substantially higher than the 45 Hz value that was initially reported for the source (Villarreal & Strohmayer 2004). The higher spin rate implies rotational broadening that would be incompatible with the narrow absorption lines initially claimed (Lin et al. 2010). Additionally, detailed model atmosphere calculations by Rauch et al. (2008) did not confirm the line identifications proposed by Cottam et al. (2002). Whether the lines are real or not remains controversial.

More recently, there have been reports of absorption edges being detected during X-ray bursts. in't Zand & Weinberg (2010) reported evidence from RXTE/PCA data for edges during so-called "superexpansion" bursts from 4U 0614+091, 4U 1722−30, and 4U 1820−30. These are bursts with powerful photospheric radius expansion (PRE) phases, that very likely drive outflows (in't Zand et al. 2012; Yu & Weinberg 2018). The edge energies were in the 5–12 keV range, and were detected after the strong, initial expansion phase. Degenaar et al. (2013) reported the detection with the Swift/XRT of a 1 keV emission line, as well as Fe-K band absorption features (lines and edges) in an energetic, intermediate-duration thermonuclear burst from the ultracompact accreting millisecond X-ray pulsar (AMXP) IGR J17062−6143 (Strohmayer et al. 2018). They suggested the 1 keV emission line could be associated with Fe-L shell transitions from the irradiation of cold gas by the burst, or alternatively, the 1s–2p Lyα transition of Ne x could also account for the observed line energy.

Kajava et al. (2017) reported evidence for absorption edges at decreasing energies between 8 and 7.6 keV in a powerful He flash from the AMXP HETE J1900.1−2455. Like in't Zand & Weinberg (2010), they argue that these features are associated with metals—the ashes of the thermonuclear burning—that have been dredged up by the vigorous convection initiated by such bursts (Weinberg et al. 2006). Kajava et al. (2017) identified the features in HETE J1900.1−2455 with the 9.28 keV photoionization edge of hydrogen-like Fe ions, but redshifted from the stellar surface to the observed energies. The variation in observed edge energy is attributed to variation in the vertical location of the photosphere. Based on this the authors estimate 1 + z ≈ 1.2, which is roughly consistent with the redshift expected from a neutron star with M = 1.4 M_⊙ and R = 12 km. Li et al. (2018) also claimed an absorption edge is present during a PRE burst from GRS 1747−312 observed with the RXTE/PCA. They reported evolution of the absorption edge energy from 9.45 to ≈6 keV through the PRE phase, and then to ≈8 keV in the cooling tail. They attributed the shift from 9.45 to 8 keV to gravitational redshift of the hydrogen-like Ni photoionization edge between the neutron star surface and the expanded photosphere during PRE. We note that this is the same burst for which in't Zand et al. (2003) reported evidence of a broad emission feature at 4.8 keV during the PRE phase, so interpretation of the apparent features in this burst as edges may not be unique.

Many of the reported detections of spectral features in X-ray bursts to date have been made with data from instruments with relatively poor resolution, such as RXTE's PCA, which has a ≈20% resolution at 6 keV. Higher-resolution detectors have typically suffered from some technical problems when observing bright X-ray bursts. The high count rates mean that CCD detectors suffer pile-up or relatively higher backgrounds are present when they must be clocked at high rates. Whereas dispersive spectroscopy with either Chandra or XMM-Newton generally mitigates any pile-up, these instruments bring relatively modest effective area to bear on the brief intervals for which bursts are bright. Fortunately, Neutron Star Interior Composition Explorer (NICER's) X-ray Timing Instrument (XTI) brings a CCD-like spectral resolution (≈80 eV at 1 keV), and substantial, pile-up-free effective area to observations of X-ray bursts. Moreover, NICER's bandpass extends down to 0.2 keV, which is ideal for observing the soft-X-ray expansion phases of PRE bursts. With the recently reported detections of edge features in such bursts, we aim to leverage NICER's unique capabilities to search for similar features.

In this paper we report results of NICER observations of thermonuclear X-ray bursts from 4U 1820−30 (hereafter, 4U 1820), which reveal both emission and absorption features during their PRE phases. 4U 1820 is a well-known, ultracompact accreting neutron star binary with an 11.4 minute orbital period. The system was one of the first discovered to produce thermonuclear X-ray bursts (Grindlay et al. 1976). Its compact nature requires a low-mass degenerate dwarf donor (Rappaport et al. 1987), and helium is likely the most abundant burst fuel (Haberl et al. 1987; Bildsten 1995; Cumming 2003).

The paper is organized as follows. We first briefly describe the observations and the general properties of the observed bursts. We then discuss our spectral analysis and the evidence for an emission line at 1 keV, and absorption lines at 1.7 and 3.0 keV in the PRE phases of these bursts. We also show that there is a systematic energy shift present in the coadded spectra of the pair of the strongest PRE bursts, compared to those in the burst pair with weaker PRE. We then discuss several scenarios for the origin of the line features, and provide possible elemental identifications. We also explore the idea that some of the observed energy shift is related to a gravitational redshift, a wind-induced blueshift, or perhaps both. We conclude with a brief summary and discussion of the implications for models of burst winds and future efforts to probe their composition, energetics, and formation.

NICER is an X-ray observatory operating on the International Space Station (ISS). It provides low-background, high-throughput (≈1900 cm² at 1.5 keV), high time resolution observations across the 0.2–12 keV X-ray band (Gendreau et al. 2012). NICER observed 4U 1820 in 2017 August, accumulating about 60 ks of exposure as part of a science team program with a major goal to detect X-ray bursts and search for spectral features in them. These observations were triggered by the entry of the source into a low flux and hard spectral state in which it is known to produce bursts (Keek et al. 2018). These data were processed using the NICERDAS (V004) software, and standard filtering and cleaning were applied. This involves excluding all data where the pointing offset is >54'', the dark Earth limb angle is <30°, the bright Earth limb angle is <40°, and the ISS was inside the South Atlantic Anomaly (SAA). We employed version v1.02 of the NICER response functions in our analysis.

2.1. Burst Light Curves

Five bursts were observed during this campaign (see Table 1). We note that detailed, time-resolved spectroscopy of burst 1 was reported in Keek et al. (2018), and it is not our goal here to reproduce such an analysis for all the observed bursts. We began by extracting light curves for each burst in several energy bands; 0.2–1 keV (soft), 2–10 keV (hard), and the full band 0.2–10 keV. Figure 1 shows the resulting soft and hard band light curves for bursts 1 (top left), 3 (top right), 4 (bottom right), and 5 (bottom left), all of which show evidence for PRE, in that the soft band rate (black curves) peaks while there is a correlated drop in the hard band rate (red curves). Burst 2 is "anomalous" in this regard. The hard band rate is actually larger than that in the soft band, and there are correlated drops in both rates during the burst rise. This behavior is suggestive of an additional absorption component present during this burst. Because it lacks a clear signature of PRE we did not consider it in the spectral analysis reported here. We will explore this interesting burst elsewhere. In Figure 2 we show the light curves of all five bursts in the full band (0.2–10 keV). The caption gives the color code to identify the bursts. We approximately aligned the rise times for display purposes so that the profiles can be compared. The cyan curve is burst 2, and the prominent dip after the initial rise is apparent.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Table 1.  X-Ray Bursts from 4U 1820 Observed by NICER

Number	ObsID	Start Time	Date	kTPRE	RPRE
		MJD (TT)	YYYY MM DD	(keV)	(km)
1	1050300108	57994.37115	2017 Aug 29	0.573 ± 0.025	103 ± 9
2	1050300108	57994.46169	2017 Aug 29	NA	NA
3	1050300109	57995.22281	2017 Aug 30	0.619 ± 0.026	102 ± 7
4	1050300109	57995.33890	2017 Aug 30	0.675 ± 0.030	78 ± 6
5	1050300109	57995.60332	2017 Aug 30	0.664 ± 0.035	72 ± 7

Note. Assigned number, observation ID, burst start time, date of observation, blackbody temperature (keV), and derived photospheric radius (km, at 8.4 kpc).

Download table as:  ASCIITypeset image

2.2. Single-burst Spectra

Guided by the results of Keek et al. (2018) we began by extracting spectra during the PRE phases for each burst. We identified the interval where the 2–10 keV band light curve peaks and extracted spectra for 0.7 s from this point for each burst. This exposure value approximately encompasses the lowest-temperature portion of the PRE phase (see Keek et al. 2018, Figure 3), and these intervals are marked in Figure 1 by the vertical dashed lines. We regrouped the spectra by combining every three successive, native energy (PI) channels, and we also applied a normalization factor derived from spectral fits to data from the Crab Nebula. See Ludlam et al. (2018) for a discussion of this procedure, which provides a way to reduce the influence of the strongest remaining unmodeled residuals in the NICER response function. We then fit each spectrum in the band 0.3–7 keV with a two-component model comprising a Comptonized component for the persistent emission and a blackbody for the thermal burst flux. We note that the NICER background is always negligible in this energy band for these bright source spectra. In XSPEC this model has the form, TBabs*(nthcomp + bbodyrad). See Życki et al. (1999) for a description of the nthcomp model.

The parameters of the nthcomp (persistent) component are qualitatively similar for all bursts. For example, for burst 3 we find, Γ = 2.2 ± 0.1, kT_bb = 0.14 ± 0.03 keV, and a normalization of 7.9 ± 1.1 (errors are 1σ). The normalization is defined as the photon flux (with units, cm⁻² s⁻¹ keV⁻¹) at 1 keV. These spectra are not sensitive to the electron temperature, kT_e, in this model, so for all fits we held it fixed at 8 keV. We left all other parameters describing the nthcomp component free to vary, but we note that the fitted parameter values describing the spectral shape are all qualitatively similar to those derived from fitting the pre-burst (persistent) emission with the same model. The derived blackbody temperature and spherical emitting radius characterizing the PRE phase do show significant variations burst to burst, and these are listed in Table 1 for each burst.

The above continuum model provides a reasonable, qualitative fit to the PRE phase spectra, but there are substantial residuals evident in the spectra. The most prominent of these is an excess near 1 keV in all of the bursts, but it is most obvious in bursts 1 and 3. As an example, we show in Figure 3 the PRE phase spectrum extracted for burst 3, where one can see excess emission above the continuum near 1 keV. Adding a Gaussian component to model this feature results in an improvement in χ² of 34.8. The fitted line centroid is 1.04 ± 0.01 keV, and an estimate of the significance, defined as the ratio of the line normalization to its 1σ uncertainty, is 6.1. The equivalent width is 47 eV. The line is unresolved in the sense that we only find an upper limit to its intrinsic width of ≈70 eV (90% confidence). The observed width, however, is consistent with NICER's measured spectral resolution at 1 keV.

Zoom In Zoom Out Reset image size

2.3. Coadded Spectra

As noted above, a similar 1 keV excess is apparent in the PRE spectra from each of the bursts, but with somewhat varying strength. Motivated by this we coadded bursts in pairs to improve the signal-to-noise ratio. In order to combine bursts that are most similar to one another, we were guided by their spectral properties inferred during the PRE phase (see Table 1). The bursts with the lowest PRE temperatures and the largest photospheric radii are bursts 1 and 3. Bursts 4 and 5 have higher temperatures and smaller radii that are consistent within the uncertainties. The burst pairs that are most similar to each other are therefore bursts 1 and 3 (pair 1), and bursts 4 and 5 (pair 2). We thus created two new spectra by coadding the single-burst spectra within each pair.

We fit the same model described above to both coadded spectra. Figure 4 shows the spectrum from pair 1, where the red curve (top) represents the best-fitting continuum model. It is consistent with the individual results for bursts 1 and 3. We find for the nthcomp component Γ = 1.98 ± 0.06, kT_bb = 0.130 ± 0.02 keV, and a normalization of 7.75 ± 0.7. For the burst blackbody component we obtain kT_PRE = 0.615 ± 0.018 keV, and R_PRE = 95.4 ± 5 km, which are also consistent with the individual results for bursts 1 and 3. We find for N_H a value of 0.17 ± 0.03 × 10²² cm⁻², which is consistent with the value reported by Güver et al. (2010).

Zoom In Zoom Out Reset image size

One can see in Figure 4 that the excess at 1 keV is now strikingly evident, and there is also now a clearer indication for a pair of absorption features near 1.7 and 3 keV. We note that the spectrum of burst 3 alone (see Figure 3) shows tentative indications for these features, but they stand out more clearly in the coadded spectrum. To assess these features further, we added three narrow Gaussians to the model. Including the line components dramatically improves the overall fit, and is shown as the cyan curve in Figure 4. Table 2 summarizes the results of the fitted line components. We find strong evidence for the presence of three lines at 1.043 ± 0.008, 1.689 ± 0.012, and 2.964 ± 0.014 keV. The strongest, at 1 keV appears in emission, and the others in absorption. As an estimate of the significance of each line we quote the ratio of the fitted line normalization to its 1σ uncertainty. All the lines are above 5σ, and the 1 keV feature is at 7σ. The bottom panel of Figure 4 shows the line residuals using the best-fitting model, but with the line normalizations set to zero. The overall spectral fit including the three line components has χ² = 279 for 227 degrees of freedom (dof), which is formally a bit high; however, the most substantial remaining residuals near 0.5–0.6 and 2.0–2.4 keV, are consistent with energy ranges with the most significant unmodeled residuals in the NICER response function (Ludlam et al. 2018). We emphasize that there are no comparably strong, nor narrow, unmodeled residuals at the centroid energies inferred for the three fitted line components. Thus, we think that a significant part of the excess χ² in our fits is likely associated with remaining uncertainties in the NICER response function. Fortunately, this will be testable, as the NICER response model is refined over time.

Table 2.  Spectral Results for Line Components

Parameter	Burst Pair 1	Burst Pair 2
Centroid Energy (keV)	1.043 ± 0.008	1.000 ± 0.008
Line Norm (ph)	0.314 ± 0.045	0.266 ± 0.045
Line width (keV)	0.06	0.07
Δχ2	61.6	44.6
σ	7.0	5.9
Equivalent width (eV)	33	27
Centroid Energy (keV)	1.689 ± 0.012	1.592 ± 0.014
Line Norm (ph)	−0.125 ± 0.023	−0.097 ± 0.016
Line width (keV)	0.07	0.07
Δχ2	29.1	22.6
σ	5.5	4.7
Equivalent width (eV)	23	21
Centroid Energy (keV)	2.964 ± 0.014	2.846 ± 0.0164
Line Norm (ph)	−0.090 ± 0.0146	−0.079 ± 0.0159
Line width (keV)	0.06	0.07
Δχ2	38.4	24.5
σ	5.9	5.0
Equivalent width (eV)	50	39

Note. Results for Gaussian line component parameters from fits to the coadded spectra of burst pairs 1 and 2. In each case three Gaussian line components have been fitted. The widths are 90% confidence upper limits, as all the lines are unresolved. The value σ is an estimate of the significance of the line, and is defined as the ratio of the line normalization to its uncertainty (1σ).

Download table as:  ASCIITypeset image

We carried out the same fitting procedure for burst pair 2, with the corresponding results summarized in Figure 5. Remarkably, there is evidence for line features near the same energies as for burst pair 1, with the emission feature near 1 keV being, again, the clearest excess (compare Figures 4 and 5). For the nthcomp continuum, in the absence of any line components, we find Γ = 1.88 ± 0.05, kT_bb = 0.123 ± 0.03 keV, and a normalization of 7.82 ± 0.6. For the burst blackbody component we obtain kT_PRE = 0.675 ± 0.026 keV, and R_PRE = 71.6 ± 5 km, which are consistent with the individual results for bursts 4 and 5. The overall fit is of a similar quality to that of burst pair 1, but with a slightly higher χ² = 295 for 227 dof. In burst pair 2 the line features appear slightly weaker. This is evident in the somewhat smaller σ values (Table 2). Nevertheless, they are all at values greater than 4.7. The fact that similar features are apparent in both independent sets of burst spectra is a very strong indication that they are real astrophysical features.

Zoom In Zoom Out Reset image size

To further explore the link between the identified line features and the PRE phase in which they are found, we extracted spectra from additional time intervals after the strong PRE phase to see if there is any evidence that the lines may persist throughout the bursts. We found no indication that the absorption lines persist beyond the strong PRE phases described above, but we found some indication for a persistence of the 1 keV emission line beyond the PRE phase in burst pair 1. As a demonstration of this we show results for spectra accumulated for 1 s intervals after the PRE phases in which the lines were identified, as well as 4 s intervals later in the decaying tail. We call these the "post-PRE" and "tail" spectra, respectively. For consistency we coadded these spectra by burst pairs in the same way as described above for the PRE phases, and we fit the same model as employed for the PRE phases. Residuals to the fits for the post-PRE (top) and tail (bottom) spectra are shown for both burst pairs in Figure 6. As is rather clear from Figure 6 (bottom), the tail spectra show no evidence for comparable line features like those seen in the PRE phases. On the other hand, for burst pair 1 there is some evidence for a persistence of the 1 keV emission feature into the post-PRE phase (see Figure 6, top). Including the 1 keV line feature to the fit in this case gives a significance of 4.2σ, measured as the ratio of the fitted line normalization to its 1σ uncertainty, and an equivalent width of 21 eV. The line centroid energy is consistent with that measured in the PRE phase. There are no comparable indications for persistence of the absorption features, and the results suggest these are rather clearly linked to the strong PRE phase. Finally, we also analyzed a 1200 s exposure of the persistent emission prior to burst 3. There are no comparable line features evident in that spectrum as well.

Zoom In Zoom Out Reset image size

Looking more closely at the measured line centroids and uncertainties, there is a very significant, systematic shift when comparing the line energies in burst pair 1 with those in pair 2 (see Table 2). All of the energy centroids measured in burst pair 2 are systematically lower. To explore this further, we computed the ratios of the line energy centroids in burst pair 1 to those in pair 2, and their uncertainties, and fit the values to a constant, s₀. The results are shown in Figure 7. The line energy ratios are statistically consistent with a constant value of s₀ = 1.046 ± 0.006, suggesting that the lines from burst pair 2 are redshifted relative to those of pair 1. Figure 8 shows the fit residuals for both coadded spectra using the best-fitting model with the line component normalizations set to zero. The top panel shows the residuals for burst pair 1 (black) and 2 (red), and the systematic energy shift is readily apparent. If we shift the residuals for pair 2 by the best-fit value of s₀ = 1.046, and then average them with those of pair 1, we obtain the "shift and add" residuals in the bottom panel.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We have presented strong evidence from NICER observations for the existence of narrow spectral lines during the PRE phases of X-ray bursts from 4U 1820. The strongest feature we identify is the emission line near 1 keV. The properties of this feature appear rather similar to the 1 keV emission line reported by Degenaar et al. (2013) in an energetic, intermediate-duration burst from the ultracompact AMXP IGR J17062−6143. 4U 1820 is now the second system to show such a line in emission, and it is intriguing that both objects are ultracompact systems likely accreting helium-rich fuel (Cumming 2003; Strohmayer et al. 2018). Interestingly, Degenaar et al. (2013) reported shifts in the line centroid energy during the IGR J17062−6143 burst that are at about the same level that we find for the 4U 1820 PRE bursts (≈1.0–1.044 keV). These authors argued that the 1 keV line could be due to Fe L shell transitions from relatively cool irradiated gas, likely in the accretion disk. They suggested that the apparent energy shifts could reflect contributions from different Fe charge states at different times during the burst. Similar processes could be at work in 4U 1820. We note that there is also the strong Lyα transition of Ne x at 1.022 keV that is in the relevant energy range, and there are several observational indications for Ne overabundances in some ultracompact binaries (Juett & Chakrabarty 2003; Strohmayer 2004). Based on this a detection of Ne in 4U 1820 would not appear to be particularly surprising. A possible scenario then is that the emission line could be produced via reprocessing (reflection) from the accretion disk, and the narrow line widths would then reflect Keplerian motions in the outer disk. Using an upper limit on the line width (HWHM) of σ_E = 1.18 × 70 = 82.6 eV, we would obtain ΔE/E ≈ v/c = 82.6/1000 = 0.0826, which implies R > 146 GM/c² ≈ 325 km for a 1.5 M_⊙ neutron star, which is not too dissimilar from the inferred maximum extent of the photosphere during PRE. This also ignores any inclination effects.

A plausible physical interpretation for the absorption lines is that they are formed in the super-Eddington wind generated by the PRE bursts. Yu & Weinberg (2018) presented hydrodynamical simulations of such winds that include modern nuclear reaction networks to follow the time-dependent burning, and composition. They find that the convective burning mixes the heavy element ashes of helium burning to sufficiently low column depths such that the ashes are eventually ejected in the wind. The ashes comprise a number of heavy elements, including; ⁴⁸Cr, ⁴⁰Ca, ⁴⁴Ti, ⁵⁶Ni, ³⁶Ar, ³²S, and ⁵²Fe. Among our detected absorption line energies, we find that for the highest energy line near 3 keV there are relatively few strong line candidates among those elements listed above. A plausible identification could be the He-like lines of S xv at 2.8839, 3.0325, and 3.1013 keV (van Hoof 2018), with one or more of these transitions being redshifted or blueshifted to the appropriate observed energy. We discuss likely sources of such shifts below. For the other absorption feature, at about 1.6–1.7 keV we note that there are plausible lines associated with several charge states of Fe, including Fe xxiii, xxiv, and xxv, as well as Cr xxiv, at about the right energy. Additionally, lines of Mg xi also fall in the appropriate range.

Alternatively, the apparent consistent energy shift between burst pairs 1 and 2 of all three spectral features seems to argue for a common origin, so we briefly comment on possible origins for the 1 keV emission line in the burst-driven wind as well. As noted by Yu & Weinberg (2018) in their hydrodynamical simulations, the base of the wind is initially at a relatively low column depth, so light (unburned) elements are launched first into the wind. As the burst progresses the base of the wind moves to greater column depths, eventually ejecting the burned ashes. Thus, composition gradients are likely to exist in the ejected wind, and perhaps emission lines could be formed in the outer parts of the wind. Another process that can form emission lines is resonant scattering (in't Zand et al. 2015), and it could be that conditions are favorable for this process to occur at some locations in the wind, such as, perhaps, in the Ne x Lyα line. More detailed theoretical calculations of radiative transfer in such burst-driven winds will be needed to explore these issues further.

A difficulty with making more precise line identifications, in addition to the need for higher spectral resolution, is the fact that both redshifts and blueshifts may be present, potentially with different contributions in the different bursts. Considering the gravitational redshift first, our spectral modeling indicates that between burst pair 1 and 2 there was an inferred variation in the average photospheric radius of ≈25 km. The difference in gravitational redshift between photons with the same rest-frame energy emitted at two different radii R₁ and R₂, expressed as the ratio of their observed energies, can be written as

$$\begin{eqnarray}&&\displaystyle \frac{{E}_{1}}{{E}_{2}}=\displaystyle \frac{{\left(1-2{GM}/{c}^{2}{R}_{1}\right)}^{1/2}}{{\left(1-2{GM}/{c}^{2}{R}_{2}\right)}^{1/2}},\end{eqnarray} \tag{ 1 }$$

where E₁ and E₂ are the observed energies of photons emitted from radii R₁ and R₂, respectively. While the continuum spectroscopy discussed above provides good evidence that the average radial location of the photosphere is different between burst pairs 1 and 2, it is likely that these values do not accurately reflect the true emission radii (Suleimanov et al. 2011). Nevertheless, if the derived radii are representative, then the observed shifts are in the correct direction, that is, the line energies for burst pair 1, with a greater photospheric radius, are higher (less redshifted) than those for pair 2, with a smaller inferred radius.

Furthermore, we can explore the assumption that all of the observed shift is due to a gravitational redshift and see what that implies for the possible values of R₁ and R₂. Figure 9 shows contours of constant line ratio, E₁/E₂, as a function of R₁ and ΔR = R₁ − R₂, for 1.5 (red) and 2.0 (blue) M_⊙ neutron stars. The red and blue contours enclose the 1σ range of our measured line ratios for 1.5 and 2.0 M_⊙ stars, respectively. The vertical and horizontal dashed lines show representative ranges (±1σ) for R₁ and ΔR, based on the inferred photospheric radii estimated from the coadded spectra of burst pairs 1 and 2. If the estimated radii are at least approximately correct, then it would appear unlikely that all of the observed shift is due to gravitational effects alone, although we note that a heavier neutron star can more easily account for the observed shift.

Zoom In Zoom Out Reset image size

Another likely culprit to induce blueshifts is the wind velocity itself. Wind velocity estimates from Yu & Weinberg (2018) suggest that the velocity is always <0.01c, which would limit a wind-induced line ratio, s_w < 1.01. Such estimates typically ignore radiative effects such as line-driving of the wind; if these are not significant, it would also appear that not all of the observed shift can be attributed to the wind. However, again an apparent shift due to this process would be in the correct direction. The stronger PRE bursts (pair 1) would be expected to have the stronger wind, and thus stronger blueshifts and higher line energies, compared to pair 2, as observed. Since both of the above processes are likely present, and work in the same direction, it seems plausible that the observed shifts are produced by a combination of both effects acting together. With the present data it appears difficult to disentangle them; however, if correlated variations in the inferred photospheric radius and line energies with time could be accurately tracked in individual bursts, then it might be possible to independently determine the gravitational redshift.

We have presented strong evidence of the presence of narrow absorption and emission lines in NICER observations of the PRE phases of X-ray bursts from 4U 1820. We find an emission line near 1 keV, and absorption features near 1.7 and 3 keV. A significant, systematic shift in the line energies between pairs of bursts with different PRE strength is strongly indicative of a real physical effect associated with the bursts, and is likely produced by a combination of gravitational redshift and Doppler blueshifts associated with the burst-driven wind. It appears plausible that all of the features are produced in the PRE wind, but reprocessing (reflection) from the accretion disk could perhaps account for the 1 keV emission line. Based on theoretical modeling of the likely composition of the PRE-generated wind, we can tentatively identify the 3 keV absorption feature with the He-like lines of S xv. The NICER data suggest a wealth of information could be gleaned from such bursts by future observations. For example, NICER observations of a longer intermediate-duration burst or superburst could perhaps track variations in the line energies as the photosphere expands, providing a way to "see" the gravitational redshift directly. In addition, future, larger collecting area instruments, such as Strobe-X, Athena and eXTP (Barcons et al. 2017; in't Zand et al. 2019; Ray et al. 2019), with the ability to observe with higher throughput and/or better spectral resolution, could provide new tools to enable more precise line identifications and perhaps use such line detections to measure the compactness of the neutron star.

This work was supported by NASA through the NICER mission and the Astrophysics Explorers Program. This research also made use of data and software provided by the High Energy Astrophysics Science Archive Research Center (HEASARC), which is a service of the Astrophysics Science Division at NASA/GSFC and the High Energy Astrophysics Division of the Smithsonian Astrophysical Observatory. D.A. acknowledges support from the Royal Society. S.G. acknowledges the support of the Centre National d'Etudes Spatiales (CNES). G.K.J. acknowledges support from the Marie Skłodowska-Curie Actions grant No. 713683 (H2020; COFUNDPostdocDTU).

Facilities: NICER - , ADS - , HEASARC. -

Software: NICERDAS (v 004), XSPEC (Arnaud 1996).