Rapidly varying ionization features in a Quasi-periodic Eruption: a homologous expansion model for the spectroscopic evolution Joheen Chakraborty, Peter Kosec, Erin Kara, Giovanni Miniutti, Riccardo Arcodia, Ehud Behar, Margherita Giustini, Lorena Hernández-García, Megan Masterson, Erwan Quintin, Claudio Ricci, Paula Sánchez-Sáez Abstract Quasi-Periodic Eruptions (QPEs) are recurring bursts of soft X-ray emission from supermassive black holes (SMBHs), which a growing class of models explains via extreme mass-ratio inspirals (EMRIs). QPEs exhibit blackbody-like emission with significant temperature evolution, but the minimal information content of their almost pure-thermal spectra has limited physical constraints. Here we study the recently discovered QPEs in ZTF19acnskyy (“Ansky”), which show absorption-like features evolving dramatically within eruptions and correlating strongly with continuum temperature and luminosity, further probing the conditions underlying the emission surface. The absorption features are well-described by dense ionized plasma of column density N H ≳ 10 21 N_{\rm H}\gtrsim 10^{21} cm -2 , blueshift 0.06 ≲ v / c ≲ 0.4 0.06\lesssim v/c\lesssim 0.4 , and either collisional or photoionization equilibrium. With high-resolution spectra, we also detect ionized blueshifted emission lines suggesting a nitrogen over-abundance of 21.7 + 18.5 − 11.0 × 21.7^{+18.5}_{-11.0}\times solar. We interpret our results with orbiter-disk collisions in an EMRI system, in which each impact drives a shock that locally heats the disk and expels X-ray emitting debris undergoing radiation pressure-driven homologous expansion. We explore an analytical toy model that links the rapid change in absorption lines to the evolution of the ionization parameter and the photosphere radius, and suggest that ∼ 10 − 3 ​ M ⊙ \sim 10^{-3}M_{\odot} ejected per eruption with expansion velocities up to v max ∼ 0.15 ​ c v_{\rm max}\sim 0.15c , can reproduce the absorption features. With these assumptions, we show a P Cygni profile in a spherical expansion geometry qualitatively matches the observed line profiles. Our work takes a first step towards extending existing physical models for QPEs to address their implications for spectral line formation. I Introduction Quasi-Periodic Eruptions (QPEs) are a newly discovered, rapidly growing class of recurring X-ray transient from supermassive black holes (SMBHs) in nearby low-mass galaxy nuclei. Thus far, they have been observed with peak luminosities of L\sim 10^{42-44} erg s -1 , recurrence times of P\sim 2.5-100\; hr, blackbody-like spectra with temperatures of kT\sim 50-250\, eV, SMBH masses of \sim 10^{5-7.5}\;M_{\odot} , and out to redshifts of z\sim 0.08 . Since their discovery in 2019, the QPE population has grown to ten [ 60 , 23 , 3 , 5 , 13 , 15 , 70 , 62 , 27 ] . A handful of potentially related phenomena are known, such as quasi-periodic oscillations in SMBHs [ 22 , 81 , 41 , 66 , 52 ] and recurring nuclear transients seen in optical [ 69 , 78 , 82 ] and X-ray [ 48 , 86 , 18 , 26 ] wavelengths. The serendipitous discovery of QPEs has inspired significant efforts to develop explanatory physical models, which separate broadly into two classes: (a) cyclic behavior related to thermal-viscous instabilities or tearing/precession of the SMBH accretion disk [ 71 , 64 , 63 , 35 , 76 , 56 ] ; or (b) interaction of the SMBH (or its accretion disk) with a \sim stellar-mass orbiting companion in an extreme mass-ratio inspiral (EMRI; Dai et al. 17 , Xian et al. 87 , Suková et al. 79 , King 37 , Zhao et al. 91 , Lu & Quataert 49 , Franchini et al. 20 , Linial & Metzger 42 , Tagawa & Haiman 80 , Kejriwal et al. 36 , Zhou et al. 93 ). The latter class of models has garnered particular attention: if QPEs are indeed the electromagnetic counterparts of EMRIs, they are the first observational probe of EMRI rates [ 6 ] and stellar dynamical phase space in external galaxy nuclei [ 73 ] , and QPEs themselves may eventually be multi-messenger sources and/or form a millihertz gravitational wave background [ 16 ] for upcoming space-based detectors (e.g. LISA , Amaro-Seoane et al. 2 ; TianQin, Luo et al. 50 ). Within the EMRI paradigm, recent activity has shifted away from scenarios which produce X-ray emission by direct accretion from the orbiting companion (e.g. Metzger et al. 55 , King 37 , Linial & Sari 46 ), instead converging towards models which produce the emission via collisional shocks of the star or black hole companion with the underlying SMBH accretion disk (e.g. Dai et al. 17 , Xian et al. 87 , Suková et al. 79 , Franchini et al. 20 , Linial & Metzger 42 , Tagawa & Haiman 80 , Zhou et al. 93 , Yao et al. 89 ), which we refer to hereafter as “orbiter-disk collision models”. There are several reasons to favor these models, of which we list three. (1) The viscous timescale at pericenter is uncomfortably long compared the several-hour timescales over which QPEs radiate their energy. Only highly eccentric EMRIs can achieve t_{\rm visc} short enough to power QPEs via accretion, but they face a fine-tuning problem to produce orbits precisely constructed to overflow some mass while avoiding tidal disruption. Furthermore, mass-overflow from a highly eccentric EMRI should produce strictly periodic events on a timescale short enough that secular effects (e.g. orbital hardening) can be neglected, which is at odds with the range of timing behaviors exhibited in QPEs. At the same time, QPE-emitting galaxies host compact accretion disks consistent with L_{X}\propto T^{4} , indicating there is a stable accretion flow whose accretion rate is linked to the eruption strength without powering them altogether [ 59 , 58 ] . Producing the emission via collisional shocks avoids the timescale problem, while still linking QPEs to the disk properties. (2) EMRI models elegantly explain the diverse range of irregularity and evolution in the burst times-of-arrival (i.e. the “Q” in QPE). Some sources show a regular alternation between two recurrence times differing by 5-10% [ 4 , 59 ] , which is naturally produced by a mildly eccentric EMRI colliding twice per orbit through ascending/descending node. Other sources show a more chaotic spread in recurrence times [ 4 , 24 ] , which have been found to show sinusoidal modulation on several-day super-periods [ 14 , 61 ] compatible with models of rigidly precessing accretion disks [ 19 , 20 ] . In two cases, QPEs tentatively show evidence for a secular period decrease , inferred from multi-year timing data of eRO-QPE2 [ 7 , 65 ] and GSN 069 [ 61 ] . This secular timing drift is a key prediction linked to the EMRI orbital decay via gravitational wave emission and viscous drag from the accretion disk [ 45 , 43 ] . (3) QPE spectral evolution is atypical compared to other nuclear transients (e.g. tidal disruption events): QPEs show hysteresis in the L-kT plane, i.e. harder/hotter bands peak before soft. For blackbody-like spectra, this can be used to infer a growing emission radius with time [ 59 , 14 , 6 ] . While shock-driven EMRI models do not uniquely explain this behavior, they provide a particularly natural physical explanation for it, best understood by analogy to the similar color-dependence in multi-band light curves of supernovae (e.g. Fig. 4 of Kasen et al. 34 ). Supernovae, like QPEs, are also thought to be powered by radiation emerging from a quasi-spherical region undergoing rapid homologous expansion following a shock breakout. Early after breakout, only the highest-energy photons can penetrate the dense, newly-compressed material. As photons continue to diffuse out, they lose energy to adiabatic losses as they do work expanding the ejecta. In this process, the ejecta become less dense and less optically thick, allowing the bulk of the photons to eventually escape by peak light. As a result, photons escaping earlier form a higher- kT spectrum (hard bands peak early and quick), and over time kT diminishes (soft bands peak late and slow). Radiative transfer simulations tracing the evolution of synthetic light curves and spectra following a orbiter-disk collision indeed find good agreement with observations of QPEs [ 83 ] . Direct tests of EMRI models are still in their infancy. Thus far, they have primarily focused on points (1) and (2) above, via modeling QPE timings with the orbital/precessional modes and evolution of the putative EMRIs [ 14 , 7 , 61 , 67 , 65 , 93 , 92 ] and fitting the properties of the underlying accretion disk to determine their compatibility with various model predictions [ 62 , 25 , 84 ] . Direct tests of the QPE radiation mechanism—point (3) —are less studied in comparison. They are complicated dramatically by the limited information content of the almost pure-thermal spectra of QPEs, making it difficult to probe the physical properties of the emission surface (e.g. expansion velocity, density, composition, ionization state). Inferring these would require detectable absorption/emission lines in addition to the continuum. This is observationally difficult, as most QPEs last only hours and have peak fluxes of \sim 10^{-13}-10^{-12} erg cm -2 s -1 ; one simply cannot collect enough photons to confidently detect anything but the continuum. The exception to this is the best-studied QPE source GSN 069, which shows evidence for persistent ionized absorption in both moderate-resolution [ 59 ] and high-resolution spectra [ 40 ] . The latter study, by stacking >1 Ms of XMM-Newton RGS data, found the first evidence for blueshifted absorption at a velocity of \sim 0.01c . The long integrations needed to generate enough signal for these detections meant they could not be probed in a phase-resolved manner, instead averaging over many bursts. The recent discovery of QPEs in ZTF19acnskyy (“Ansky”; Hernández-García et al. 27 ), a 10^{6}\;M_{\odot} SMBH at redshift z=0.024 with a newborn accretion flow [ 74 ] , provides an exciting opportunity to overcome the observational barriers laid out above. With the longest burst durations/peak fluxes by factors of 4/2.5\times respectively, Ansky accumulates 10 \times more counts per eruption than any other known QPE source. The high flux is due to the relatively large peak luminosity ( L_{\rm peak}\sim 3\times 10^{43} erg s -1 ) and the relatively nearby redshift, though Ansky is neither the closest nor the most luminous QPE. [ 27 ] reported the first detection of a highly significant absorption-like spectral feature evolving rapidly within individual bursts, never seen before in a QPE. In particular, recent work on orbiter-disk collisions suggest that the QPE emission may arise from a shock-heated, quasi-spherical debris cloud ejected from the accretion disk, which subsequently undergoes free homologous expansion [ 42 , 83 ] . In this manuscript, we analyze the properties of such an expansion, and whether they can effectively reproduce the spectral evolution observed in Ansky. In Section II , we present the data products along with phenomenological fits to the spectra. Details of the observations and data analysis procedures can be found in Appendix A . In Section III , we fit collisional ionization and photoionization plasma models to the moderate- and high-resolution spectra, then discuss our interpretation of this feature within the orbiter-disk collision framework. In Section IV we make concluding remarks and highlight directions for future study of QPE emission mechanisms. Figure 1: Left: light curve from a subset of the total 60-day monitoring campaign of Ansky. Top right: moderate-resolution spectra at the highlighted phases from NICER (red) and XMM-Newton EPIC-pn (blue) fit with a blackbody model (solid lines). Middle and bottom right: We separated the NICER and XMM spectra, and plot the ratios of the data to the best-fitting blackbody model; all spectra show a significant deficit of counts compared to the model between 0.6-1.2 keV. The residuals are well-described by an additional gaussian absorption component ( gabs ), which are shown by the solid lines. II Results Figure 2: Left: Time evolution of the centroid line energy of the gaussian absorption component. We overplot the flare profiles for comparison. All NICER spectra used for these measurements are shown in Fig. B.1 . Right: “phase-folded” version of the left panel, i.e. with t-t_{\rm peak} on the x -axis. The flare profiles are also normalized to peak flux. Figure 3: Correlations of the blackbody temperature ( kT ), luminosity ( L_{X} ), and various parameters of gaussian absorption component: centroid ( \mu_{E} ), width ( \sigma ), and strength ( A_{E} ). We note that the gabs model in XSPEC defines the line strength such that the optical depth at the line center is \tau=A_{E}/\sigma/\sqrt{2\pi} . In each plot, we only include the points for which the error calculation converged for both relevant parameters. In the legends we include the best linear fits. QPEs predominantly show blackbody-like X-ray spectra with characteristic temperatures of 50-250 eV [ 60 , 23 , 3 , 13 ] . They also have a quiescence component well-described by a disk blackbody ( T_{\rm in}\sim 50 eV) which does not vary as dramatically, suggesting an underlying compact accretion flow. The only observation taken during quiescence with sufficient sensitivity for a detection was a 25.1 ks XMM-Newton exposure (OBSID 935191501); fitting with a disk blackbody model, we estimate a quiescence flux of F_{X}=2\times 10^{-14} erg cm -2 s -1 ( L_{X}=3\times 10^{40} erg s -1 ), which is a factor of \sim 500 fainter than the typical QPE peak [ 27 ] . Due to the extreme peak-to-quiescence amplitude in this source, we neglected the disk blackbody component for our fits, instead just using a blackbody while accounting for Galactic ISM absorption with a neutral column density fixed to N_{\rm H}=2.6\times 10^{20}\;{\rm cm}^{-2} [ 28 ] . The faint quiescence and short exposure limits us from searching for discrete spectral features between the eruptions, and the remainder of our analysis focuses on data taken during the eruptions. We performed spectral fitting in XSPEC [ 8 ] , using a model of tbabs \times zbbody (with the blackbody redshifted to the host, z=0.024 ). The resulting fits show significant absorption-like residuals in the extreme Wien tail, spanning characteristic energies of \sim 0.6-1.2 keV with a strong dependence on QPE phase, as reported in [ 27 ] . We show four sample spectra from different QPEs and different phases in Fig. 1 , where the data-to-model ratios being systematically lower than one illustrates the inadequacy of a blackbody-only model. The residuals are seen in NICER , XMM EPIC-pn, MOS1, MOS2, and RGS spectra, confirming their astrophysical origin. The Wien tail residuals appear even for more complicated choices of continuum model: we tested bulk motion Comptonization ( bmc ), a disk blackbody ( diskbb ), and two-component continua comprising a disk plus a blackbody, a double-blackbody, and a blackbody plus Comptonization. None of these choices are able to explain the high-energy residuals. To explore the time evolution of this feature, we divided the data into eight time bins per QPE defined by 0-25/25-50/50-75/75-100% of peak flux and vice versa for the decline (see Appendix A.1 for details), then performed phase-resolved fits of the absorption feature with a gaussian profile. We found that a single gaussian absorption line is almost always sufficient to account for the residuals in moderate-resolution spectra, while a gaussian emission line cannot describe the residuals well. We show ratio plots (data divided by blackbody model) from all epochs in which the data quality was sufficient to detect or exclude the presence of this feature in Appendix B . The median 0.3–2 keV C-stat/d.o.f. is 1.54 without the gaussian absorption component, and 1.09 with it. For the median degrees of freedom 104 without the gaussian line, this corresponds to a median fit improvement of \Delta C-stat = 50 for three additional parameters (line centroid, strength, and width). In a handful of spectra, there is an excess emission component red-ward of the gaussian absorption line akin to a P Cygni profile (e.g. QPE 5 peak in Fig. 1 ), but generally it is not significant enough to appreciably alter the fit statistic. With phenomenological fits in hand, in Fig. 2 we show the time-evolution of the gaussian centroid energy across QPEs. A recurring overall trend is visible: typically the line energy starts at \sim 0.7-0.8 keV early in the rise phase of the QPEs, then increases to a maximum of \sim 1-1.2 keV coinciding with peak flux, then finally decreases during the QPE decline phase back down to \sim 0.6-0.7 keV. The left panel shows that the higher-luminosity QPEs tentatively coincide with higher-energy lines, though we do not have the data quality to robustly confirm this claim. We do not show the same time-evolution plots for the line strengths and widths, because they are far more poorly constrained due to degeneracies with the background models and low total counts. In general however, they also show a similar time-evolution as the centroid energy, and the blackbody temperature and luminosity are clearly correlated with the line centroid, strength, and width. In Fig. 3 , we show the empirical relations between various parameters of the gaussian absorption model evolving as functions of time, as well as linear fits for each pair of variables. We note that our analysis is not sensitive to features potentially appearing at lower fluxes or during quiescence, because fainter bins have insufficient counts to confirm the presence (or absence) of the gaussian lines. We thus do not comment on whether the absorption disappears altogether at low fluxes, only focusing on the detections at higher L . III Discussion III.1 Physical modeling with ionized absorption/emission Figure 4: Data-to-blackbody ratios for the XMM EPIC-pn spectrum of OBSID 0935191601 (QPE #7 decline), and six different physical model fits of dense, ionized, blueshifted absorbing plasma (Table 1 ). All models are statistically near-identical, despite traversing a huge range in bulk velocity ( 0.06\lesssim v/c\lesssim 0.4 ), broadening ( 0\lesssim v_{\rm rms}\lesssim.058c ), and ionization parameter/electron temperature. Table 1: XMM EPIC-pn spectral fitting results with photoionization and collisional ionization equilibrium plasma in SPEX . The baseline model (1) is a redshifted blackbody with Galactic ISM absorption. Models (2-4) add photoionized blueshifted plasma absorption ( HOT \times REDS \times PION \times BB ), while models (5-7) assume collisional ionization equilibrium ( HOT \times REDS \times HOT \times BB ). We found three statistically comparable solution classes with low (2 & 5), medium (3 & 6), and high (4 & 7) blueshifts. Model Parameter Value Unit \pm 1\sigma Error C-stat/d.o.f. (1) Blackbody Temperature ( T_{\text{BB}} ) 67.4 eV +0.17/-0.17 247/168 Area ( 4\pi R_{\rm BB}^{2} ) 1.96\times 10^{24} cm 2 (+3.5/-3.5) \times 10^{22} (2) Blackbody Temperature ( T_{\rm BB} ) 70.0 eV +0.23/-0.30 176/164 Area ( 4\pi R_{\rm BB}^{2} ) 1.73\times 10^{24} cm 2 (+6.4/-5.9) \times 10^{22} Photoionized plasma absorption ( PION ): low- v_{z} solution Column density ( N_{\rm H} ) 6.04\times 10^{22} cm -2 (+4.0/-1.4) \times 10^{20} Ionization ( \log\xi ) 4.18 erg s -1 cm +0.24/-0.16 RMS velocity ( \sigma ) 331 km s -1 +115/-107 Bulk velocity ( v_{z} ) -1.48\times 10^{4} km s -1 (+4.8/-3.9) \times 10^{3} (3) Blackbody Temperature ( T_{\rm BB} ) 68.9 eV +0.23/-0.30 177/164 Area ( 4\pi R_{\rm BB}^{2} ) 1.96\times 10^{24} cm 2 (+0.70/-1.0) \times 10^{23} Photoionized plasma absorption ( PION ): medium- v_{z} solution Column density ( N_{\rm H} ) 4.01\times 10^{21} cm -2 (+5.8/-5.3) \times 10^{20} Ionization ( \log\xi ) 1.97 erg s -1 cm +0.31/-0.17 RMS velocity ( \sigma ) 121 km s -1 +77/-38 Bulk velocity ( v_{z} ) -4.70\times 10^{4} km s -1 (+7.8/-0.81) \times 10^{3} (4) Blackbody Temperature ( T_{\rm BB} ) 69.1 eV +0.61/-0.41 175/164 Area ( 4\pi R_{\rm BB}^{2} ) 1.82\times 10^{24} cm 2 (+3.9/-4.4) \times 10^{22} Photoionized plasma absorption ( PION ): high- v_{z} solution Column density ( N_{\rm H} ) 1.58\times 10^{21} cm -2 (+3.9/-2.4) \times 10^{20} Ionization ( \log\xi ) 2.75 erg s -1 cm +0.37/-0.23 RMS velocity ( \sigma ) 1.75\times 10^{4} km s -1 (+5.3/-4.5) \times 10^{3} Bulk velocity ( v_{z} ) -1.31\times 10^{5} km s -1 (+2.3/-5.0) \times 10^{3} (5) Blackbody Temperature ( T_{\rm BB} ) 69.2 eV +0.27/-0.37 177/164 Area ( 4\pi R_{\rm BB}^{2} ) 1.83\times 10^{24} cm 2 (+4.6/-4.1) \times 10^{22} Collisionally ionized plasma absorption ( HOT ): low- v_{z} solution Column density ( N_{\rm H} ) 3.85\times 10^{22} cm -2 (+1.8/-0.89) \times 10^{22} Electron temperature ( T_{e} ) 397 eV +32/-21 RMS velocity ( \sigma ) 187 km s -1 +167/-77 Bulk velocity ( v_{z} ) -1.78\times 10^{4} km s -1 +390/-450 (6) Blackbody Temperature ( T_{\rm BB} ) 69.1 eV +0.29/-0.29 175/164 Area ( 4\pi R_{\rm BB}^{2} ) 2.25\times 10^{24} cm 2 (+1.1/-0.93) \times 10^{23} Collisionally ionized plasma absorption ( HOT ): medium- v_{z} solution Column density ( N_{\rm H} ) 4.41\times 10^{21} cm -2 (+6.6/-6.1) \times 10^{20} Electron temperature ( T_{e} ) 115 eV +6.2/-5.4 RMS velocity ( \sigma ) 399 km s -1 +180/-150 Bulk velocity ( v_{z} ) -5.07\times 10^{4} km s -1 (+4.8/-1.0) \times 10^{3} (7) Blackbody Temperature ( T_{\rm BB} ) 69.2 eV +0.37/-0.44 176/164 Area ( 4\pi R_{\rm BB}^{2} ) 1.74\times 10^{24} cm 2 (+2.4/-0.55) \times 10^{23} Collisionally ionized plasma absorption ( HOT ): high- v_{z} solution Column density ( N_{\rm H} ) 4.03\times 10^{21} cm -2 (+1.5/-2.5) \times 10^{20} Electron temperature ( T_{e} ) 213 eV +23/-109 RMS velocity ( \sigma ) 1.61\times 10^{4} km s -1 (+5.0/-5.5) \times 10^{3} Bulk velocity ( v_{z} ) -1.21\times 10^{5} km s -1 (+0.76/-1.5) \times 10^{4} Table 2: XMM RGS spectral fitting results using photoionized plasma models. We first tested a blackbody with cosmological redshift and ISM absorption (model 1). We then used a photoionized emission model ( HOT \times REDS \times PION \times BB ; model 2) to capture the discrete features from \sim 0.5-0.7 keV. The RGS spectrum is consistent with the Wien-tail absorption residuals seen in EPIC-pn, but not with sufficient SNR to constrain models, so we do not include it. In Fig. 5 we showed a composite model (not a fit) consisting of the emission-line plasma below, and one of the absorption components of Table 1 . Model Parameter Value Unit \pm 1\sigma Error C-stat/d.o.f. (1) Blackbody Temperature ( T_{\text{BB}} ) 73.0 eV +0.55/-0.50 1688/1408 Area ( 4\pi R_{\rm BB}^{2} ) 1.05\times 10^{24} cm 2 (+5.4/-5.4) \times 10^{22} (2) Blackbody Temperature ( T_{\rm BB} ) 70.1 eV +1.4/-1.5 1650/1403 Area ( 4\pi R_{\rm BB}^{2} ) 1.26\times 10^{24} cm 2 (+1.2/-1.1) \times 10^{23} Photoionized plasma ( PION ) emission Column density ( N_{\rm H} ) 2.44\times 10^{21} cm -2 (+1.2/-1.1) \times 10^{21} Ionization ( \log\xi ) 2.85 erg s -1 cm +0.26/-0.29 RMS velocity ( \sigma ) 1.03\times 10^{4} km s -1 (+1.9/-2.1) \times 10^{3} Bulk velocity ( v_{z} ) -1.17\times 10^{4} km s -1 (+1.9/-2.6) \times 10^{3} N abundance 21.7 solar +18.5/-11.0 Figure 5: Simultaneous spectra from XMM RGS 1+2 (crosses) and EPIC-pn (squares). In the middle panel, we show the ratio of the data to a blackbody-only model. We mark lines and edges of ionized O/N/Fe with the corresponding blueshifts found by model (2) of Table 2 . Broad features upon a thermal continuum have been seen in the X-ray spectra of nuclear transients before. In the literature, it is common to interpret them as blueshifted ionized emission and/or absorption from a relativistic outflow or accretion disk wind. For instance, similar spectral features have been reported in a handful of tidal disruption events, including ASASSN-14li [ 33 , 1 ] , 1ES 1927+654 [ 72 , 51 ] , AT2022lri [ 90 ] , ASASSN-20qc [ 39 , 68 ] , AT2020ksf [ 85 ] , and even the QPE source GSN 069 [ 40 ] . In all of those sources, the line features are seen between \sim 0.7-1.5 keV. In contrast to Ansky, their properties do not vary rapidly, changing in strength/centroid only over months to years rather than just kiloseconds. It is thus unclear whether they are driven by the same underlying phenomenon, and whether using the same models to describe them is truly appropriate. Nevertheless, to allow comparison with previous results, we performed fits with photoionized and collisionally ionized plasma. Such models are generally best constrained by high-resolution spectra, which we take advantage of here. We performed these fits with the SPEX X-ray spectral fitting code [ 32 ] due to its built-in capability to perform photoionization calculations self-consistently while varying the illuminating spectrum, using the PION model due to [ 54 ] . We first fit only the EPIC-pn data; the resulting solutions are reported in Table 1 . We began by fitting a baseline model of a redshifted blackbody, accounting for Galactic ISM absorption with a neutral plasma (the HOT model with temperature fixed to 10^{-6} K) with N_{\rm H}=2.6\times 10^{20} cm -2 fixed (a total model of HOT \times REDS \times BB ). We then modeled the Wien-tail absorption features with photoionized (models 2-4) and collisionially ionized (models 5-7) blueshifted absorbing plasmas, i.e. total models of HOT \times REDS \times PION \times BB and HOT \times REDS \times HOT \times BB respectively. We found three solution classes spanning a wide range of blueshifts, around v\approx-0.06c (models 2 and 5), v\approx-0.16c (models 3 and 6) and v\approx-0.4c (models 4 and 7). All absorption models (2-7) result in significant fit improvements of \Delta C-stat =70-72 . It is remarkable that, with the resolution of EPIC-pn, the fits are all statistically near-identical despite the vastly different physical parameters. We plot the change in the model resulting from the addition of absorbing plasma in Fig. 4 ; the similarity of the models when down-sampled to CCD resolution is manifest. In Fig. 5 , we show the high-resolution RGS spectrum taken simultaneously with EPIC-pn from ObsID 0935191601, which occurred just after QPE peak and captured most of the decline phase of QPE #7. The RGS spectrum is consistent with the Wien tail residuals of EPIC-pn, but has neither the signal nor the energy coverage to constrain absorption models at those energies. However, the RGS data also reveal narrower emission features at lower energies, further suggesting the presence of blueshifted high-ionization lines of O/N. We therefore proceeded by freezing the absorbing outflow parameters to those found by with EPIC-pn, and adding a lower-velocity emission component in a two-zone plasma model. We used the absorption plasma solution from model 3 of Table 1 , because it is the median v_{z} solution, though we checked that the fit parameters are insensitive to this choice because the absorption features occur predominantly at energies which RGS does not probe. In contrast to the features in EPIC-pn, which are equally well-described by collisional or photoionization, only a photoionized plasma can effectively capture the emission lines seen in RGS. We report the resulting parameters with PION model fits in Table 2 ; the addition of the emission-line plasma results in a significant fit improvement of \Delta C-stat = 38. Another noteworthy result from our high-resolution spectroscopy is a large required nitrogen over-abundance of \sim 21.7\times solar. Significant CNO processing is also seen in GSN 069—which is the only other QPE with robust measurements [ 75 , 40 ] —as well as in many tidal disruption events from both ultraviolet and X-ray spectroscopy [ 12 , 38 , 88 , 57 ] . Before proceeding, we again stress that our absorption fits are not unique, as the EPIC-pn energy resolution cannot definitively constrain the outflow velocity, density, ionization state, broadening or ionization process. Doing so would require high-resolution spectroscopy, with high enough fluxes/long enough exposures to generate enough signal to capture the absorption features well into the Wien tail. The purpose of our fits is to illustrate the presence of dense ionized plasma with relativistic outflow velocities, as has been seen in the literature before. Our fits do not readily provide an explanation for the time-evolution for the features, in particular the rise-and-decay of the outflow velocity one would infer from Fig. 2 . III.2 Relation to EMRI models In the previous section, we have shown that the broad absorption features in our data can be described equally well by collisionally ionized or photoionized equilibrium plasmas, with a huge possible range of blueshifts 0.06\lesssim v/c\lesssim 0.4 . All models face some difficulties: for example, those with low/medium blueshifts (2-3 and 5-6) require unusually low velocity broadening compared to their bulk velocities, while the models with high blueshifts (4 and 7) require extreme energetics. Without direct high-resolution spectroscopic constraints on the absorption profiles, it is difficult to make further observational progress and exclude or favor a particular solution. We therefore turn to arguments based on theoretical grounds, and discuss the characteristic blueshifts, broadening, and ionization states we expect may be produced within the emerging orbiter-disk collision paradigm invoked for QPEs. We refer the reader to Fig. 1 of Vurm et al. 83 for a relevant physical picture of the QPE emission source. Each orbiter-collision results in a highly supersonic radiation-mediated shock, in which a cross-section of debris from the accretion disk is quickly raised to higher temperature, pressure, and energy density than the surrounding disk material. This over-pressurized debris breaks out of the disk midplane, proceeding to expand quasi-spherically due to radiation pressure from the photons generated in the early phases following the shock. It is the photon generation and subsequent expansion of this ejected disk material, rather than the star or the disk themselves, which is proposed to power the QPE emission. In this case, the absorption line blueshift is set by the debris expansion velocity rather than the bulk velocity of the ejected material. Modeling the time-evolving properties of the ionization features with an accretion disk wind or relativistic outflow, rather than a homologous expansion, would be an interesting exercise, but is outside the scope of this work. Over the course of the ejecta expansion, the outer layers experience a significant luminosity from the inwardly propagating photon diffusion surface, resulting in significant ionizing flux throughout the entire eruption duration. We thus expect photoionization processes to dominate over collisional ionization, as the latter contributes only during the much briefer orbiter-disk impact duration. However, we emphasize that this is a geometry- and model-dependent assumption. If a spherical homologous expansion is not the correct picture for QPEs in Ansky, one could reduce the contribution of photoionization processes, in which case Models (5-7) of Table 1 may be the more appropriate ones to consider. Nevertheless, for the remainder of the discussion we focus on interpreting our results within the homologous expansion picture, and defer further considerations of collisional ionization-dominated ejecta to future work. Previous studies have examined the evolution and energetics of the debris via analytical arguments [ 42 , 20 , 43 ] , as well as hydrodynamical and Monte Carlo simulations [ 89 , 83 ] . We proceed by briefly describing the key properties of these models, directly considering their implications for spectral line formation, and comparing qualitatively to the features observed in our data. III.3 Energetics of the shock breakout and homologous expansion phase The total energy budget powering each QPE, only a fraction of which escapes as radiation, is set by the dissipated orbital energy during each collision ( E_{\rm tot} ). Initially, this is entirely deposited in the form of internal energy ( E_{\rm int} ), that is, heat plus the energy of trapped radiation. Part of this internal energy is subsequently radiated away as photons powering the observed eruptions ( E_{\rm rad} ), while a larger portion goes into adiabatic losses driving the ejecta expansion (and is thus converted into bulk kinetic energy, E_{\rm kin} ). The total energy at any given time following the orbiter-disk collision is therefore partitioned among: E_{\rm tot}=E_{\rm rad}+E_{\rm kin}+E_{\rm int}. At early times, we have E_{\rm tot,0}=E_{\rm int,0} , and as the eruption progresses, we quickly tend towards E_{\rm int}\ll E_{\rm rad}t_{\rm peak}\end{cases} (8) where L_{\rm peak} is the peak luminosity reached at a time t_{\rm peak} , and \tau_{1} / \tau_{2} are the e -folding times for the rise/decay. Figure 6: Physical parameters of our analytical toy model for the ejecta evolution. As described in Section III.3 , we assume the QPE is powered by photons generated by the orbiter-disk collision. We follow the evolution of the escaping overdense ejecta as it is quickly accelerated by radiation pressure, and then reaches passive homologous expansion at a time t_{0} as light begins to escape (as detailed in Vurm et al. 83 ). The ejecta mass is radially dispersed over the course of the \sim 10^{5} sec eruption as the internal energy endowed by the orbiter-disk collision is converted to bulk kinetic energy driving the expansion, converging to E_{\rm kin}/E_{\rm rad}\sim 3 . Once homologous expansion is reached, the photosphere recedes deeper into the ejecta over time, until the ejecta eventually becomes transparent. The corresponding velocity, v_{\rm ph}/v_{\rm max}=R_{\rm ph}/R_{\rm max} , is characterized an initial rise as the ejecta is accelerated, then a decline during the retreat of R_{\rm ph} . This two-phase evolution of \log\bar{\xi} and v_{\rm ph} provide a compelling explanation for the spectral line shifts seen in Fig. 2 . Having laid out the arguments above, we now use them to construct a toy model for the ejecta evolution over a single QPE of Ansky. We assumed that the debris initially experiences a constant acceleration over a few radial doubling times, then quickly reaches passive expansion for the remaining evolution. For the homologous expansion phase, we adopt fiducial values of p_{1}=1 , p_{2}=10 , R_{0}=10^{11} cm, R_{\rm br}(t)=0.5R_{\rm max}(t) (chosen to maintain similarity with Fig. 12 of Vurm et al. 83 ), M_{\rm ej}=10^{-3}M_{\odot} , and \mu m_{p}=0.6m_{\rm H} . We are pushed to these relatively large ejecta masses to obtain a long enough photon diffusion timescale to reproduce the extreme burst durations of Ansky. Though we adopt R_{0}=10^{11} cm, our modeling is basically insensitive to any choice of R_{0}\lesssim 10^{13} cm, as we find expansion velocities reaching v_{\rm max}\sim 0.15c , meaning the ejecta will traverse 10^{13} cm within just \sim 2 ks, a small fraction of the typical eruption durations; at this scale, it makes little difference whether the debris started from R_{0}=0.1R_{\odot} , R_{\odot} or 10R_{\odot} . For our model light curve, we use L_{\rm peak}=1.5\times 10^{43} erg s -1 , \tau_{1}=20 ks, \tau_{2} = 30 ks, and t_{\rm peak}=50 ks, motivated by the typical profile of the QPEs in Ansky. We plot the evolution of six relevant physical parameters in our toy model— L(t) , \rho(t,r) , v(t,r) , \bar{\xi}(t) , R_{\rm ph}(t) , and v_{\rm ph}(t) —in Fig. 6 . For our adopted values, Eq. 2 yields a density: \displaystyle\rho_{0}(t) \displaystyle=\frac{M_{\rm ej}}{4\pi}\bigg(\frac{575R_{\rm max}^{3}(t)}{7168}-\frac{R_{0}^{2}R_{\rm max}(t)}{4}\bigg)^{-1} \displaystyle\approx 1.9\times 10^{-11}\frac{\rm g}{\rm cm^{3}}\bigg(\frac{t}{10^{4}\;\rm s}\bigg)^{-3}\bigg(\frac{v_{\rm max}}{0.15c}\bigg)^{-3}\bigg(\frac{M_{\rm ej}}{10^{-3}M_{\odot}}\bigg) (9) where the last line is an approximation which is true by the time the ejecta has reached homologous expansion ( R_{0}\ll R_{\rm max} ). We can also write an expression for the velocity of the debris outer edge by combining Eqs. 2 - 3 : \displaystyle v_{\rm max} \displaystyle=\sqrt{\frac{2E_{\rm kin}}{M_{\rm ej}}}\bigg(\frac{8960R_{0}^{2}R_{\rm max}^{2}-2875R_{\rm max}^{4}}{4480R_{0}^{4}-497R_{\rm max}^{4}}\bigg)^{1/2} \displaystyle\approx\sqrt{\frac{11.6E_{\rm kin}}{M_{\rm ej}}}\qquad(R_{0}\ll R_{\rm max}) (10) While we cannot directly determine E_{\rm kin} , E_{\rm rad} may be used as a proxy, as it is a measurable quantity from the light curve. Assuming isotropic emission, Ansky radiates an average \sim 10^{48} erg integrated per flare, roughly a factor of 10 higher than the shorter-period QPEs [ 27 ] . [ 83 ] found via Monte Carlo radiative transfer simulations that by peak light, QPE ejecta converge roughly to E_{\rm kin}/E_{\rm rad}\sim 3 . So, we can express Eq. 10 as: \displaystyle v_{\rm max} \displaystyle\approx 0.15c\bigg(\frac{E_{\rm rad}}{10^{48}\;\rm erg}\bigg)^{1/2}\bigg(\frac{E_{\rm kin}/E_{\rm rad}}{3}\bigg)^{1/2}\bigg(\frac{M_{\rm ej}}{10^{-3}M_{\odot}}\bigg)^{-1/2}. (11) We note that in the left panel of Fig. 2 , we tentatively see that higher-luminosity eruptions are indeed associated with higher-energy line centroids, suggesting that v_{\rm max} indeed scales with E_{\rm rad} . The v_{\rm max} we infer from Eq. 11 and Table 1 imply significantly larger debris velocities than that inferred from the change in R_{\rm bb}(t) as measured from fitting the X-ray spectra, which imply an expansion of only \sim 1.5\times over the course of the eruption [ 27 ] This discrepancy, however, has been noted before in QPEs, as the inefficiency of photon production and the presence of an electron-scattering atmosphere result in the measured R_{\rm bb} significantly underestimating any true physical length scale or expansion velocity within the debris (for details, see Vurm et al. 83 ). Conveniently, our choice of parameters results in R_{\rm ph}>R_{\rm br} for most of the burst duration. Eq. 6 therefore gives: \displaystyle\frac{R_{\rm ph}}{R_{\rm max}}(t) \displaystyle=\Big(\frac{\kappa_{\rm es}\rho_{0}R_{\rm max}(t)}{6144+\kappa_{\rm es}\rho_{0}R_{\rm max}(t)}\Big)^{1/9} \displaystyle\approx 0.7\times\bigg(\frac{t}{10^{4}\;\rm s}\bigg)^{-2/9}\bigg(\frac{v_{\rm max}}{0.15c}\bigg)^{-2/9}\bigg(\frac{M_{\rm ej}}{10^{-3}M_{\odot}}\bigg)^{1/9} (12) Recalling that R_{\rm ph}/R_{\rm max}=v_{\rm ph}/v_{\rm max} , we use Eqs. 9 and 12 to write the photosphere velocity over time: v_{\rm ph}(t)\approx 0.11c\times\bigg(\frac{t}{10^{4}\;\rm s}\bigg)^{-2/9}\bigg(\frac{v_{\rm max}}{0.15c}\bigg)^{7/9}\bigg(\frac{M_{\rm ej}}{10^{-3}M_{\odot}}\bigg)^{1/9} (13) Plugging Eq. 12 for the photosphere velocity into Eq. 7 also allows us to approximate the time-evolution of the density-weighted ionization parameter: \displaystyle\bar{\xi}(t) \displaystyle\approx 9.7\times 10^{4}\;\rm erg\;\rm s^{-1}\;\rm cm \displaystyle\times\bigg(\frac{L(t)}{10^{43}\;\rm erg}\bigg)\bigg(\frac{\mu m_{p}}{0.6m_{\rm H}}\bigg)\bigg(\frac{t}{10^{4}\;\rm s}\bigg)^{-1}\bigg(\frac{v_{\rm max}}{0.15c}\bigg)^{-1} \displaystyle\times\bigg[1.43\bigg(\frac{t}{10^{4}\;\rm s}\bigg)^{2/9}\bigg(\frac{v_{\rm max}}{0.15c}\bigg)^{2/9}\bigg(\frac{M_{\rm ej}}{10^{-3}M_{\odot}}\bigg)^{-1/9}-1\bigg] (14) We reiterate that Eqs. 12 - 14 rely on the convenient fact that R_{\rm ph}\geq R_{\rm br} for nearly the entire time-evolution of our toy model, so that we did not have to treat the lower case of Eq. 6 . Perturbations from our chosen parameters will quickly run into models in which the photosphere retreat is faster (e.g. for lower M_{\rm ej} ), and then Eqs. 12 - 14 become poor approximations. We defer more general-case modeling, which can explore full expressions for the late-time evolution and/or relax the assumptions on the assumed density profile, to future work. We plot numerical integrations of \log\bar{\xi} , R_{\rm ph}/R_{\rm max} , and v_{\rm ph}/c following Eq. 5 across 100 timesteps, along with the analytic expressions provided by Eqs. 12 - 13 , in the bottom middle and right panels of Fig. 6 . In our toy model, the two-phase evolution of the line profiles in Fig. 2 arises due to a combination of the rise and decline in \log\bar{\xi} , changing the relevant line species and ionization balance contributing to the ionization features, while v_{\rm ph} simultaneously decreases, reducing over time the blueshifts at which they appear. We comment that our toy model outlined above is pushed to somewhat extreme values for the ejecta mass. Maintaining a particle density large enough to have \log\bar{\xi}\lesssim 4 requires M_{\rm ej}\sim 10^{-3}M_{\odot} , in contrast to earlier studies which typically assume M_{\rm ej}\sim 10^{-5}-10^{-4}M_{\odot} [ 42 , 83 ] . Such low masses would result in \log\xi\gg 5 , at which point all ions relevant for soft X-ray transitions will be fully stripped of electrons and contribute no absorption. The underlying reason is the extremely long burst durations of Ansky: with such a large r\sim vt , the characteristic \xi=\frac{L}{nr^{2}}\propto Lr (Eq. 9 ) is pushed to very high values due to the linear distance scaling. This large radial expansion of the photosphere, by \sim 50\times over the entire eruption duration, also implies there should be significant adiabatic cooling of the ejecta—by a corresponding 0.02\times —whereas the blackbody temperatures only appear to decrease by a factor \sim 0.5\times [ 27 ] . It is unclear how this 25\times deviation from adiabatic cooling is achieved. One possibility is very inefficient photon production throughout the entire burst, which would result in fewer photons sharing the same (decreasing) internal energy and cause the emission temperature to appear substantially higher than in matter-radiation equilibrium [ 46 , 83 ] . However, our toy model requires a more dramatic inefficiency than invoked in prior studies, again owing to the unusually large expansion radii. On its face, solving this would require v_{\rm max} smaller by \sim 25\times ; however, such a small velocity would be totally incapable of explaining the broad emission lines seen in our data. With a characteristic \sim 4\times 10^{48} erg released per burst, each QPE dissipates a nontrivial fraction of the total orbital energy available to the EMRI, E_{\rm orb}=GM_{\rm BH}m/(2a) . A 1 M_{\odot} EMRI has E_{\rm orb}\sim 1.6\times 10^{51} erg, i.e. within 400 eruptions ( \sim 2000 days) the entire orbital energy would be depleted as the system is promptly driven to merger. A higher-mass EMRI companion would resolve the immediacy of this issue, as with E_{\rm orb}\gtrsim 10^{53} erg the orbit possesses enough energy to remain stable for > thousands of orbits. In any case, the dissipation of orbital energy in powering the QPEs appears to be much faster than the gravitational wave timescale; the QPEs in Ansky were first detected in early 2024, so we may expect to see significant evolution in the QPE properties and timescales within just a couple of years. Observations will quickly weigh on this prediction. We also draw attention to the fact that the large ejecta masses result in a much longer-lived photosphere retreat than in typical QPE models. In the bottom-middle panel of Fig. 6 , we see that by the end of the 120 ks eruption duration, R_{\rm ph}\sim 0.4R_{\rm max} , i.e. a 40% of the debris has yet to become transparent. We suggest that the further retreat of the photosphere may be accompanied by photons experiencing such large adiabatic losses that they are observed in ultraviolet rather than X-rays, a possibility also suggested in [ 44 ] and [ 83 ] . Remarkably, Ansky shows hints of rapid UV variability unusual for a QPE [ 74 , 27 ] , though not with sufficiently high cadence or significance to resolve any relation to the X-ray bursts. Follow-up higher-cadence UV monitoring has the potential to reveal whether the X-ray QPEs in this source are indeed accompanied by a longer-lived UV tail. III.4 Line formation in expanding QPE atmospheres Existing models for absorption by a photoionized/collisionally ionized plasma, such as the PION and HOT models used in Section III.1 (but also the commonly-used XSTAR model within XSPEC ) are designed to compute the transmission and absorption through a uniform slab geometry of ionized plasma. This is not appropriate for a spherical homologous expansion experiencing a significant density gradient of several orders of magnitude. Therefore, keeping with our assumptions for the debris density profile and geometry, we turn to the problem of spectral line formation in the expanding spherical QPE atmospheres. Line formation in spherically expanding media has been explored in detail by [ 77 ] , [ 11 ] , [ 34 ] , and others. We describe the effect of the rapidly expanding debris cloud on a simple delta-function line profile as a means of illustrating the qualitative expectation. For much of the following argument, we follow [ 11 ] , albeit with some key differences with respect to the line resonance condition and profile. Figure 7: Top panels: schematic of spherically symmetric debris in homologous expansion (the apparent deviation from spherical geometry is due to the nonlinear axes, which correspond to the velocity blueshift at each z -coordinate). We indicate contours of both density and radial coordinate r , and mark the z -coordinate corresponding to the line-of-sight velocity on the top axis. The observer sits at z=-\infty , with a sight-line peering through the debris until it becomes occluded by the photosphere at R_{\rm ph} . Bottom panels: line profiles corresponding to each choice of debris parameters, along with observational data from NICER / XMM . They are not fits to the data, and are overplotted only to show the qualitative agreement with the model. Relying on the assumption of spherical symmetry, we work in a two-dimensional polar coordinate system ( p , z ) centered on the ejecta, which is expanding radially in all directions (independent of polar/azimuthal angle). We orient the observer along the z -axis at -\infty , so that we only consider light rays traveling along -\hat{z} . The coordinate axes are shown in the top panels of Fig. 7 ; this definition of z coincides conveniently with the observed velocity blueshift, owing to the homologous expansion assumption ( r\propto v_{r} , z\propto v_{z} ). The cylindrical coordinates ( p , z ) are related to the usual polar coordinates ( r , \theta ) by r^{2}=p^{2}+z^{2} and z=-r\cos\theta . The radiative transfer equation in this coordinate system reads: \displaystyle\frac{dI(z,p,\nu)}{dz} \displaystyle=\rho(r)\kappa(\nu,r)[-S(r)+I(\nu,p,z)] \displaystyle=n(r)\sigma(\nu,r)[-B_{\nu}(T(r))+I(\nu,p,z)] (15) where we have assumed the source function in the atmosphere is an isotropic blackbody spectrum depending only on the radial coordinate within the ejecta. To solve this equation for the intensity I(\nu,p,z) , we divide the debris into a discrete boundary at the photosphere, and assume the photosphere radiates a single-temperature blackbody intensity I_{\rm ph}=B_{\nu}(T_{\rm ph}) independent of spatial coordinate within the boundary R_{\rm ph} . The solution to the radiative transfer equation in the expanding envelope ( r>R_{\rm ph} ) is thus divided into the contributions from the near and far sides of the photosphere: \displaystyle I(\nu,p,z) \displaystyle=\int_{n\sigma z}^{\infty}S\big(\sqrt{p^{2}+z^{\prime 2}}\big)e^{\tau(\nu,p,z)-\tau^{\prime}(\nu,p,z^{\prime})}d\tau^{\prime}(\nu,p,z^{\prime}) \displaystyle\qquad\qquad\qquad\qquad\mathrm{if}\;p>R_{\rm ph}\ \mathrm{or}\ z>0 \displaystyle I(\nu,p,z) \displaystyle=\int_{n\sigma z}^{-n\sigma\sqrt{R_{\rm ph}^{2}-p^{2}}}S\big(\sqrt{p^{2}+z^{\prime 2}}\big)e^{\tau(\nu,p,z)-\tau^{\prime}(\nu,p,z^{\prime})}d\tau^{\prime} \displaystyle\qquad\qquad+B_{\nu}(T_{\rm ph})\exp\Bigg(-\int_{z}^{-\sqrt{R_{\rm ph}^{2}-p^{2}}}n\sigma dz^{\prime}\Bigg) \displaystyle\qquad\qquad\qquad\qquad\mathrm{if}\;pR_{\rm ph}\ \mathrm{or}\ z>0 \displaystyle I(\nu,p,-\infty) \displaystyle=B_{\nu}(T_{0})\Big(1-e^{-n(r_{0})\sigma cR_{\rm max}/v_{\rm max}}\Big) \displaystyle\qquad\qquad+B_{\nu}(T_{\rm ph})e^{-n(r_{0})\sigma cR_{\rm max}/v_{\rm max}} \displaystyle\qquad\qquad\qquad\mathrm{if}\;p