Solar flares, filament eruptions and coronal mass ejections (CMEs) are thought to be caused by a common magnetohydrodynamic process, though not all of them are necessarily observed in the same event. Magnetic reconnection is a key energy release mechanism for flares, which are thought to be sometimes triggered by the instability of cool filaments in active regions1. Recently, it has been discussed that much larger ‘superflares’ that release the energy of more than 1033 erg (ten times the largest solar flares, ~1032 erg) can occur—or have occurred relatively recently—even on the Sun3,4,5,8. Superflares may produce much larger CMEs than the largest solar flares, which can severely affect the environment, habitability and development of life around young and intermediate-age stars6. However, superflares on solar-type stars have been mainly detected by optical photometry (for example, the Kepler space telescope)3. Therefore, no observational indication of filament eruptions/CMEs has been reported for solar-type stars. Optical spectroscopic observations are a promising way to detect stellar filament eruptions, which can be indirect evidence of CMEs. However, for solar-type stars, optical spectra of superflares have never been obtained.

EK Draconis (EK Dra) is known to be an active young solar-type star (a G-type, zero-age main-sequence star with an effective temperature of 5,560–5,700 K and age of 50–125 Myr; ref. 9) that exhibits frequent UV stellar flares10,11 and gigantic starspots at low–high latitudes9. We conducted optical spectroscopic monitoring of EK Dra for 19 nights between 21 January 2020 and 15 April 2020, simultaneously with optical photometry from the Transiting Exoplanet Survey Satellite (TESS)12. Time-resolved neutral-hydrogen Hα-line spectra at 6,562.8 Å (radiation from cool plasma of a few times 10,000 K) were spectroscopically observed at the 3.8 m Seimei telescope13 and the 2 m Nayuta telescope. In this campaign, we succeeded in obtaining optical spectra of large superflares on a solar-type star. The superflare that occurred on 5 April 2020 was simultaneously observed using TESS photometry in white light (~6,000–10,000 Å) and ground-based spectroscopy in the Hα line (Fig. 1a,b and Extended Data Fig. 1). The Hα brightening was associated with the TESS white-light flare, which lasted 16 ± 2 min. The radiated bolometric energy of the TESS white-light flare is estimated to be 2.0 ± 0.1 × 1033 erg (20 times the most energetic solar flares), and the radiated Hα-line energy was 1.7 ± 0.1 × 1031 erg; thus, the flare is classified as a superflare.

Fig. 1: Light curves and spectra of a superflare on EK Dra.
figure 1

a, The light curve observed by TESS in white light (~6,000–10,000 Å) on BJD (barycentric Julian day) 2458945.2 (5 April 2020). The individual points represent the stellar flux normalized by the averaged value with the pre-flare level subtracted. The 1σ value of the pre-flare light curve (−150 min to 0 min) is shown in blue. b, Light curves of the Hα EW observed using the medium-dispersion spectrograph MALLS (Medium and Low-Dispersion Long-Slit Spectrograph) at the Nayuta telescope (grey circles) and the low-dispersion spectrograph KOOLS-IFU (Kyoto Okayama Optical Low-Dispersion Spectrograph with optical-fibre Integral Field Unit) installed at the Seimei telescope (red triangles) during the same observing period as in a. The Hα emissions were integrated within ±10 Å from the Hα line centre (6,562.8 Å) after dividing by the continuum level, and the pre-flare level was subtracted. The positive and negative values represent emission and absorption, respectively, compared with the pre-flare level. The 1σ value of the pre-flare light curve (−150 min to 0 min) is plotted in red and black for Seimei and Nayuta data, respectively. c,d, Two-dimensional Hα spectra obtained using the Seimei telescope (c) and the Nayuta telescope (d). The red and blue colours correspond to emission and absorption, respectively. The dashed lines indicate the stellar surface gravity (g*) and half of the surface gravity (0.5 g*). c and d share the upper colour bar. e, Temporal evolution of the pre-flare-subtracted Hα spectra observed using the Seimei telescope (red) and the Nayuta telescope (black), with the spectra shifted by constant values for clarity. The spectra are binned in time, and the integration periods correspond to the horizontal axes of ad. The intensities are normalized by the stellar continuum level. The vertical dotted line indicates the Hα line centre, and the horizontal dotted lines indicate the zero levels for each spectrum. The 1σ error bar around the line core, based on the residual scattering in the line wing, is also shown.

Source data

After the impulsive phase, the TESS white-light intensity returned to its pre-flare level. However, the equivalent width (hereafter EW) of Hα (the wavelength-integrated Hα emission normalized by the continuum level) became lower than the pre-flare level (that is, it displayed enhanced absorption), returning to the pre-flare level after approximately 2 h (Fig. 1b). The blueshifted Hα absorption component with a maximum central velocity of about −510 km s−1 and a half-width of ±220 km s−1 appeared soon after the superflare. The velocity gradually slowed with time, and a redshifted absorption component appeared at a few times 10 km s−1 (Fig. 1c–e and Extended Data Figs. 2a and 3a). Both ground-based spectroscopic observations simultaneously recorded the same spectral change, demonstrating that low-temperature and high-density neutral plasma above the stellar disk moves at high speed toward the observer before some parts finally start to fall back to the surface. In addition, the deceleration is not monotonic: it was 0.34 ± 0.04 km s−2 in the initial phase, dropping to 0.016 ± 0.008 km s−2 in the later phase (Fig. 1c,d and Extended Data Fig. 3b). This is interpreted in terms of changes in the height of the ejected mass. The observed deceleration is in good agreement with that due to the surface gravity of approximately 0.30 ± 0.05 km s−2 (ref. 9), although the initial value is slightly larger.

How much do the stellar spectral changes obtained here actually resemble those of solar filament eruptions? Blueshifted Hα absorption profiles are often observed from solar filament eruptions1,14. As in Fig. 2, we generated spatially integrated Hα spectra of a solar flare/filament eruption that occurred on the solar disk using the SMART (Solar Magnetic Activity Research Telescope) data15 (Extended Data Fig. 4 and Supplementary Video 1). We converted to the full-disk pre-flare-subtracted spectra by multiplying by the partial-region/full-disk ratio (that is, virtual Sun-as-a-star spectra). We found that the blueshifted absorption component at approximately 100 km s−1 was predominant soon after the solar flare, and the spatially integrated Hα EW showed enhanced absorption (Fig. 2a). These blueshifted profiles are unequivocally due to the filament eruption. Later, the blueshifted component decelerated and gradually turned into slow, redshifted absorption (Fig. 2b,c). The Hα EW returned to the pre-flare level in approximately 40 min (Fig. 2a). Although the energy scales and velocities are different, the solar data strongly resemble the spectral changes in the superflare on EK Dra (see Supplementary Information for another event). This similarity suggests that the stellar phenomenon is the same as the simply magnified picture of the solar filament eruption.

Fig. 2: The space-integrated light curves and spectra of a C5.1-class solar flare and filament eruption on 7 July 2016, observed with the SDDI (Solar Dynamics Doppler Imager) installed at SMART.
figure 2

a, GOES (Geostationary Operational Environmental Satellite) soft X-ray (1–8 Å) and Hα EW light curves of the solar flare are plotted as a blue dashed line and red circles, respectively. The Hα emissions were integrated within ±1.5 Å from the Hα line centre (6,562.8 Å) and were divided by the full-disk continuum level, and the pre-flare level was subtracted. Time 0 is the time when the flare begins. b, Two-dimensional pre-flare-subtracted Hα spectra. The red and blue colours correspond to emission and absorption compared with the pre-flare levels, respectively. The dashed line indicates gravity at the solar surface. c, Temporal evolution of the pre-flare-subtracted Hα spectrum shifted vertically by constant values for clarity. The Hα spectra were produced by integrating the data over a large enough region to cover the flaring area (Extended Data Fig. 4). The intensities are normalized by the total solar continuum level. The vertical dotted line indicates the Hα line centre, and the horizontal dotted lines indicate the zero levels for each spectrum.

Source data

A filament eruption is the only explanation for the blueshifted absorption component on EK Dra by solar analogy1. The hypothesis that the blueshifted absorption on EK Dra might come from up-/downflow in flare kernels must be rejected because they never show Hα absorption16,17. Also, downflow in cooled magnetic loops (known as post-flare loops)14 shows redshifted absorption, so they cannot explain the blueshifted absorption. (However, the redshifted absorption in EK Dra in the later phase might be caused by post-flare loops14.) Rotational visibilities of prominences or spots also are not adequate to explain it, since the rotation speed of EK Dra is only 16.4 ± 0.1 km s−1 (ref. 9). Thus, we concluded that we detected a stellar filament eruption on the solar-type star.

Some observational signatures for stellar filament eruptions or CMEs have been reported previously for cooler K–M dwarfs18,19,20,21,22 and evolved giant stars23 (see Methods and refs. 6,24 for reviews). The observation of a giant star shows a blueshifted X-ray emission line of 90 km s−1 in the post-flare phase and hotter CME is proposed as a possible explanation23. Recently, X-ray/extreme UV dimmings have been reported as indirect evidence of stellar CMEs on K–M dwarfs22. In M-dwarf flares, many blueshifted Balmer/UV line emission components have been reported18,19,20,21,24, which are interpreted as filament eruptions. Some M-dwarf flares share properties with the eruption on EK Dra: the blueshift emissions have high velocities of hundreds of kilometres per second, and some exhibit velocity changes and appear after the impulsive phase20,21. For M-dwarf events, the number of studies reporting highly time-resolved velocity variations of blueshift components is still insignificant (~5 min cadence), and a simultaneous white-light flare has never been detected. Our detection of a stellar filament eruption is reliable because we provided solar counterparts, highly time-resolved spectra (~50 s cadence) and a simultaneous TESS white-light flare.

What properties does the filament eruption on EK Dra have? The maximum observed velocity of the blueshifted component was ~−510 km s−1 with a width of 220 km s−1. This is larger than the typical velocities of solar filament eruptions (10–400 km s−1) associated with CMEs2, although it is a little smaller than the escape velocity at the surface on EK Dra (~670 km s−1). The cool plasma reached at least ~1.0 stellar radius from the stellar surface (or the initial height) as derived by integrating the velocity over time (or ~3.2 stellar radii from the stellar surface on the basis of the deceleration rates). In this case, a projection angle of at most 45° can be allowed when we assume that the event occurs on the disk centre. On this projection angle, the velocity can be up to ~−720 km s−1, so there is a possibility that the velocities of some components of the EK Dra eruption could exceed the escape velocity. However, it should be noted that there are weak redshifted components with a velocity of a few 10 km s−1 in the late phase, indicating that some materials fell back to the star. This is often observed in the case of solar filament eruptions with CMEs25.

The filament area is estimated to be 1.6 × 1021 cm2 (5.6% of the stellar disk), and the erupted mass is calculated to be 1.1\({}_{-0.9}^{+4.2}\times 1{0}^{18}\) g on the basis of the absorption components. The mass is more than ten times larger than those of the largest solar CMEs26,27 (it should be noted that the mass can be somewhat under-/overestimated; Methods). This mass estimate is in reasonable agreement with those predicted from empirical26,27 and theoretical28 solar scaling relations between CME mass and flare energy within the error bars (~9.4\({}_{-2.4}^{+3.2}\times 1{0}^{16}\) and 3.1\({}_{-1.1}^{+1.6}\times 1{0}^{17}\) g for refs. 27 and 26, respectively) (Fig. 3a). This suggests that the stellar filament eruption can share a common underlying mechanism with smaller-scale filament eruptions/CMEs (that is, magnetic energy release1,28) although the absolute values of most physical quantities are very different.

Fig. 3: Mass and kinetic energy as a function of flare energy for solar and stellar flares and filament eruptions/CMEs.
figure 3

a, Comparison between bolometric flare energy and ejected mass. The red square represents the superflare on EK Dra, the black crosses denote solar CME data, the green triangles signify data for solar prominence/filament eruptions and surges taken from previous studies and the green plus sign is the solar filament eruption/surges displayed in Fig. 2 and Supplementary Fig. 9 (Velocity, mass, and kinetic energy: solar data), respectively (Table 1). Note that solar ‘surges’ are jet-like filament eruption phenomena (see Another case of solar flares on 2 April 2017 for an explanation of the surge). The cyan dashed and magenta dotted lines are trend fits for solar CMEs expressed as MCMEsE0.59 and MCMEsE0.7, respectively (Solar flare energy–CME mass relation and refs. 26,27). b, Comparison between flare bolometric energy and kinetic energy of the erupted mass. The symbols are the same as in a. The cyan dashed line is a fit for solar CMEs expressed as EkinEX1.05. The kinetic energy of eruption on EK Dra is calculated to be 3.5\({}_{-3.0}^{+14.0}\times 1{0}^{32}\) erg, which is outside the error range of the predicted value of 4.8\({}_{-0.9}^{+1.1}\times 1{0}^{33}\) erg (ref. 27). The error bars are derived as the model errors (see ‘Velocity, mass and kinetic energy: stellar data’).

Moreover, the kinetic energy is calculated to be 3.5\({}_{-3.0}^{+14.0}\times 1{0}^{32}\) erg, which is 16% of the radiation energy in white light. The magnetic energy stored around the starspots on EK Dra can be at least 8.0 × 1035 erg, which is enough to produce superflares and filament eruptions with energy of ~1033 erg. In addition, this value is slightly smaller than those extrapolated from the solar CME scaling law (4.8\({}_{-0.9}^{+1.1}\times 1{0}^{33}\) erg; ref. 27) (Fig. 3b), which is similar to the filament eruption/CME candidates on other stars24. In previous studies, it has been argued that kinetic energy can be reduced by overlying magnetic fields24,29. The deceleration of our events was a few times 10% larger than the stellar gravity (Extended Data Fig. 3b). The strong magnetic fields on EK Dra have been reported before9 and may support the above explanations. However, its small kinetic energy can also be understood through a solar analogy: the velocities of (lower-lying) filament eruptions are usually four to eight times lower than those of the corresponding (higher-lying) CMEs2, and therefore the kinetic energies of filament eruptions are typically smaller (green symbols in Fig. 3b).

Did a CME occur in this event? Obviously, the line-of-sight velocity of ~510 km s−1 was lower than the escape velocity and some masses fell back, which may indicate a so-called ‘failed’ filament eruption29. However, this does not necessarily mean that a CME did not occur, again by solar analogy. In fact, the erupted filaments often fall back to the Sun when CMEs happen. For example, a well studied solar event on 7 June 2011 involved a 200–600 km s−1 filament eruption where much filamentary material fell back to the Sun, but some mass clearly escaped as a CME with velocities of ~1,000 km s−1 (ref. 25 and Supplementary Information). The event on EK Dra may correspond to this solar event. In addition, ref. 30 showed that whether a solar filament eruption leads to a CME can be simply distinguished by a parameter of (Vr_max/100 km s−1)(L/100 Mm)0.96, where Vr_max is the maximum radial velocity and L is the length scale (Fig. 4). When the parameter is more than ~0.8, the probability that a filament eruption leads to a CME is more than 90% (ref. 30). The value of the parameter of eruption on EK Dra is ~18, meaning that our detection of the fast and sizable stellar filament eruption is indirect evidence that mass escapes into interplanetary space as a CME.

Fig. 4: Statistical properties of solar filament eruptions taken from ref. 30 and comparison with the stellar filament eruption on EK Dra.
figure 4

Comparison between the Vr_max and L of filament eruptions. The blue circles and orange crosses indicate the solar filament eruptions with and without CMEs, respectively. The red square corresponds to the stellar filament eruption on EK Dra obtained here (see Stellar velocity, mass and kinetic energy data for the calculation of the length scale). Note that the stellar data are the observed line-of-sight velocities. Since the deceleration of the stellar filament eruption corresponds well to the surface gravity and the absorption component of the filament is visible on the disk all the time, we can expect that the filament is flying in our direction perpendicularly to some extent, so there would not be such a large difference between radial velocity and line-of-sight velocity. We expect that the radial velocity can be larger than the line-of-sight velocity if we assume the projection effect, while it will be about √2 times smaller at most if it erupts at a 45° tilt in the radial direction, which does not change our discussion. The solid line indicates the threshold that can roughly distinguish filament eruption with and without CMEs derived in ref. 30. The threshold can be expressed as (Vr_max/100 km s−1)(L/100 Mm)0.96 = 0.8, which is determined by using the Linear Support Vector Classification algorithm (see ref. 30 for the detailed method).

Finally, we summarize future directions of our findings (see Supplementary Information for details): It is speculated that the filament eruptions/CMEs associated with superflares can severely affect planetary atmospheres6. Our findings can therefore provide a proxy for the possible enormous filament eruptions on young solar-type stars and the Sun, which would enable us to evaluate the effects on the ancient, young Solar System planets and the Earth, respectively. Further, it is also speculated that stellar mass loss due to filament eruptions/CMEs can affect the evolutionary theory of stellar mass, angular momentum and luminosity7,26 more importantly than can stellar winds. At present, frequency and statistical properties of CMEs on solar-type stars are unknown, but important insights into these factors will be obtained by increasing the number of samples in the future.


TESS light-curve analysis

TESS observed EK Dra (TIC 159613900) in its sectors 14–16 (18 July 2019–6 October 2020) and 21–23 (21 January 2020–15 April 2020). The TESS light curve from the 2 min time-cadence photometry was processed by the Science Processing Operations Center pipeline, a descendant of the Kepler mission pipeline based at the NASA Ames Research Center12,31. Extended Data Fig. 1 shows the light curve of EK Dra from BJD 2458945 (JD 2458944.997, 5 April 2020 11:56 ut; Sector 23), and the stellar superflare detected by TESS, the Seimei telescope and the Nayuta telescope in Fig. 1 is indicated with the red arrow in this figure. The quasiperiodic brightness variation is thought to be caused by the rotation of EK Dra with the asymmetrically spotted hemisphere3,5. The rotation period is reported as about 2.8 d (ref. 9). Although the superflare occurred near the local brightness maximum, some of the starspots are expected to be visible from the observer32,33,34,35. In Extended Data Fig. 1, other flares are also indicated using black arrows with more than two consecutive observational points whose flaring amplitude is more than three times the TESS photometric errors3,36. The white-light flare energy was calculated by assuming the 10,000 K blackbody spectra36,37 (Flare energy). The pixel-level data analysis is shown in TESS pixel-level data analysis. The estimated occurrence frequency of superflares (>1033 erg) in the TESS band was about once per 2 d, which means that about 12 nights’ monitoring observations are necessary on average to detect one superflare from the ground-based telescope under a clear-sky ratio of 50%. This implies that our datasets are highly unique.

Spectroscopic data analysis

Here, we present the utilization of low-resolution spectroscopic data from KOOLS-IFU38 of the 3.8 m Seimei telescope13 at Okayama Observatory of Kyoto University and MALLS19,39 of the 2 m Nayuta telescope at Nishi-Harima Astronomical Observatory of the University of Hyogo. KOOLS-IFU is an optical spectrograph with a spectral resolution of R (λλ) ~ 2,000 covering a wavelength range from 5,800 to 8,000 Å; it is equipped with Ne gas emission lines for wavelength calibration and instrument characterization. The exposure time was set to be 30 s for this night. The sky spectrum was subtracted by using the sky fibres for each spectrum. The data reduction follows the prescription in ref. 40. During this observation, the signal-to-noise ratio (S/N) for one frame is typically 172 ± 6. The observations using the Seimei telescope ended just after 133.7 min (Fig. 1b–d).

MALLS is an optical spectrograph with R ~ 10,000 at the Hα line covering a wavelength range from 6,350 to 6,800 Å; it is also equipped with Fe, Ne and Ar gas emission lines for wavelength calibration and instrument characterization. The sky spectrum was subtracted using a nearby region along the slit direction for each observation. The exposure time was set to be 3 min for this night. The MALLS data reduction follows the prescription in ref. 19. The S/N for one frame is typically 86 ± 8 during this observation. For the MALLS data, the wavelength corrections are also performed for each spectrum using the Earth’s atmospheric absorption lines.

We corrected the wavelength for the proper motion velocity of −20.7 km s−1 of EK Dra on the basis of Gaia Data Release 2 (ref. 41). Continuum levels are defined by a linear fit between the wavelength ranges of the Hα line wing (6,517.8–6,537.8 and 6,587.8–6,607.8 Å). We take the continuum level as the wavelength range between 6,517.8–6,537.8 and 6,587.8–6,607.8 Å to measure the EW (=∫(1 − Fλ/F0) dλ, where F0 is the continuum intensity on either side of the absorption feature, while Fλ represents the intensity across the entire wavelength range of interest). The original spectra are shown in Stability of pre-flare spectra. Extended Data Fig. 2 shows the pre-flare-subtracted Hα spectra during and after the superflare on EK Dra with higher time cadence than Fig. 1e. The narrowband Hα EW (Hα − 10 Å–Hα + 10 Å) is used for the measurements of the radiated energy and duration of the Hα flare because of the high S/N, and the broadband Hα EW (Hα − 20 Å–Hα + 10 Å) is used for the measurements of the amount of absorption (that is, mass and kinetic energy).

Solar data analysis

In the main text, we showed the data of a C5.1-class solar flare (that is, the peak GOES soft X-ray flux FGOES is 5.1 × 10−6 W m−2; hereafter ‘Event 1’, see Table 1) and associated filament eruption around 07:56 ut, 7 July 2016, observed using the SDDI15 installed on the SMART at Hida Observatory (further explanations continue to Solar data analysis (continued from Methods)). The SDDI conducted a monitoring observation of the Sun on 7 July 2016. It takes full-disk solar images at 41 wavelength points every 0.5 Å from the Hα line centre − 8.0 Å (−366 km s−1) to the Hα line centre + 8.0 Å (366 km s−1), while it takes images every 0.25 Å from the Hα line centre − 2.0 Å (−91 km s−1) to the Hα line centre + 2.0 Å (91 km s−1). Each set of images is obtained with a time cadence of 20 s and a pixel size of about 1.2 arcsec. The SDDI started the daily monitoring observations in 2016, and the C5.1-class solar flare is one of the largest solar flares with a filament eruption among the events observed by SDDI with good weather conditions in these five years. The solar filament eruption was also reported in refs. 15,42. Another jet-like filament eruption (known as a solar surge1) associated with a C8.0-class solar flare is also shown in Another case of solar flares on 2 April 2017 (hereafter we call this surge ‘Event 2’, see Table 1).

Table 1 Properties of solar filament eruptions/surges reported in previous studies and this study. The data are plotted in Fig. 3. The calculation of flare energy, mass and kinetic energy is introduced in Velocity, mass, and kinetic energy: solar data

This paper used 70 min time series of the SDDI images taken from 07:30 ut on 7 July 2016 (Supplementary Video 1). As in Extended Data Fig. 4, the C5.1-class flare occurred around an active region, named ‘NOAA 12561’, on the solar disk, and was accompanied by a typical filament eruption15,42. The spectra from the event are integrated over a spatial region that is large enough to cover the visible phenomena (the magenta region in Extended Data Fig. 4a,b). The spectra are reconstructed by using the template solar Hα spectrum convolved with the SDDI instrumental profile.

Here, we define L(λ, t, A) as the luminosity at a wavelength of λ and time of t that is integrated for the region A (that is, L(λ, t, A) = ∫AI(t) dA; I(t) is intensity). We now define Alocal as the integration region (magenta region in Extended Data Fig. 4a,b), and Afull-disk as the solar full disk. We first obtain the local (partial-image) pre-flare-subtracted spectra ΔSlocal, which are normalized by the local (partial-image) total continuum level (L(6,570.8 Å, t, Alocal)):

$${{\Delta }}{S}_{{{{\rm{local}}}}}=\frac{L(\lambda ,t,{A}_{{{{\rm{local}}}}})-L(\lambda ,{t}_{0},{A}_{{{{\rm{local}}}}})}{L(6{,}570.8\,\mathring{\rm A} ,t,{A}_{{{{\rm{local}}}}})},$$

where t0 is a given time of the pre-flare period. Then, the (virtual) full-disk pre-flare-subtracted spectra ΔSfull-disk are obtained by multiplying by the ratio of the partial-image continuum to full-disk continuum (total continuum ratio):

$${{\Delta }}{S}_{{{{\rm{full}}}}{\mbox{-}}{{{\rm{disk}}}}}={{\Delta }}{S}_{{{{\rm{local}}}}}\frac{L(6{,}570.8\,{\mathring{\rm{A}}} ,{t}_{0},{A}_{{{{\rm{local}}}}})}{L(6{,}570.8\,{\mathring{\rm{A}}} ,{t}_{0},{A}_{{{{\rm{full}}}}{\mbox{-}}{{{\rm{disk}}}}})},$$

and we obtain a virtual pre-flare-subtracted spectrum of this phenomenon as if we observed the Sun as a star. The EW of the Hα is also calculated using the ΔSfull-disk, and we obtained the virtual Sun-as-a-star ΔHα EW (that is, differential Hα flux normalized by the full-disk continuum level).

Stellar velocity, mass and kinetic energy data

For the stellar filament eruption, the velocity is derived by fitting the absorption spectra obtained using the Seimei telescope with the normal distribution N(λ, μ, σ2), where μ is the mean wavelength and σ2 is the variance. In Extended Data Fig. 3a, we plotted the temporal evolution of the velocity ((μ − λ)/λc, where λ is 6,562.8 Å and c is light speed) for the fitted absorption feature with the width of σ. We only plotted the data whose absorption features are clear enough to fit the shape with the threshold of the fitted absorption amplitude of >0.01 and fitted velocity dispersion of <500 km s−1 and >100 km s−1. The threshold was determined by trial and error, and we find that many missed detections of absorption features occur when we select threshold values other than this one. The amplitude value of 0.01 corresponds to the detection limit when considering the typical S/N ~ 170 of the Seimei telescope/KOOLS-IFU, and the lower limit of 100 km s−1 is determined to avoid detecting the sharp noisy signals. About 27% of data points were discarded due to this threshold from the initial points (22 min) to final points (110 min), especially for the latter decaying phase. Here, the maximum observed velocity and its errors are calculated as 510 ± 120 km s−1 with a width of 220 ± 90 km s−1 from the mean values of the μ and σ of the first five points (t = 22–26 min in Fig. 1), respectively. The mean value of the velocity when the absorption becomes strong (t = 25–50 min in Fig. 1) is estimated as 258 km s−1.

The plasma mass is simply calculated from the total Hα EW. We used the simple Becker cloud model43 with optical depth at the line centre of the ejected plasma τ0 of 5 (which is slightly more optically thick than solar filament eruptions; compare ref. 44), two-dimensional aspect ratio of 1 (that is, cubic), local plasma dispersion velocity W of 20 km s−1 and source function S of 0.1 on the basis of the solar observations45. The observed half-width of 220 km s−1 of the stellar blueshifted component is larger by one order of magnitude than the solar value, but here we use the solar value as a template. The dispersion velocity of 220 km s−1 is considered to be the upper limit of the local velocity dispersion because the ejected mass would have a complex two-dimensional velocity distribution, which can cause larger W in the integrated spectra. First, the modelled EW of enhanced absorption is calculated by using the Becker cloud model, when the plasma velocity vshift is −258 km s−1, as

$${{{\rm{model}}}}\ {{{\rm{EW}}}}={\int}_{\lambda }\frac{{I}_{\lambda }-{I}_{0\lambda }}{{I}_{0,{{{\rm{Cont}}}}.}}\,{\mathrm{d}}\lambda ={\int}_{\lambda }\frac{S-{I}_{0\lambda }}{{I}_{0,{{{\rm{Cont}}}}.}}(1-{\mathrm{e}}^{-{\tau }_{\lambda }})\,{\mathrm{d}}\lambda$$
$${\tau }_{\lambda }={\tau }_{0}\exp \left[-\frac{1}{2}{\left(\frac{\lambda /{\lambda }_{0}-(1+{v}_{{{{\rm{shift}}}}}/c)}{W/c}\right)}^{2}\right],$$

where I0λ is background intensity and I0,Cont. is continuum intensity. This is the EW value for an extreme case when the full disk of the star is completely covered with absorbing, cool ejected plasma. By comparing the modelled EW (equation (3)) with the lowest observed stellar EW value of −0.16 Å (integrated for Hα − 20 Å–Hα + 10 Å; Supplementary Fig. 4c), the cool-plasma filling factor is calculated to be 5.9% of the stellar disk (that is, modelled EW/observed EW; area = 1.6 × 1021 cm2). Using the length scale of the ejected plasma, 3.9 × 1010 cm (=area0.5), the hydrogen column density is derived as 4.0 × 1020 cm−2 from the assumed optical depth based on the plasma model46. In the model of ref. 46, hydrogen/electron density is calculated by assuming an ionization equilibrium for a population of hydrogen atoms due to a balance between recombination and radiative photoionization through the Balmer/Lyman continuum. It should be noted that the ionization equilibrium of filaments on active stars may be somewhat different from the solar observations due to their high UV radiations, which may affect the evaluation of the mass of the ejecta. By multiplying the hydrogen column density by the filament area, we then obtained the plasma mass of 1.1 × 1018 g. If the two-dimensional aspect ratio becomes 0.1, similar to a jet-like feature (x width:y width:z depth = 1:0.1:0.1), then the estimated mass becomes larger by a factor of 1.78. If optical depth ranges from 0.8 to 10 (ref. 44), the source function takes values of 0.02 or 0.5 and the dispersion velocity is 10 or 220 km s−1 (ref. 45), then the estimated masses change by a factor between 0.15 and 4.9. In Fig. 3a, we used the mass of 1.1\({}_{-0.9}^{+4.2}\times 1{0}^{18}\) g for an optical depth of 5, and the uncertainties of the model (0.15–4.9) are used as the error bars since the model-based errors are expected to be much larger than the observational errors. It should be noted that this mass estimate could be either a significant overestimate of the mass of an affiliated CME due to most of the filament falling back to the star, or a significant underestimate due to most of the CME actually being hot coronal material rather than cool filament. The plasma kinetic energy is then calculated as 3.5\({}_{-3.0}^{+14.0}\times 1{0}^{32}\) erg by using the velocity of 258 km s−1. The observed maximum velocity was 510 km s−1 in the early phase, so the kinetic energy can be larger by a factor of 4 although the absorption component was weak at that time.

Related works on candidate stellar filament eruptions or CMEs on other types of star

Here, we discuss potential stellar filament eruptions/CMEs reported in the previous studies (see refs. 6,24,47 for reviews). In other stars, such as M-type stars18,19,20,21,21,48,49,50,51,52,53,54,55,56, K-type stars57, T Tauri stars58,59, close binaries60,61 and giant stars23, some observational candidates for stellar filament eruptions/CMEs have been reported, although confirmations of filament eruptions/CMEs in analogy with solar observations are still rare. We also note that some other studies have tried to detect a signature of stellar filament eruptions/CMEs in various ways but have not succeeded in robust detection62,63,64,65,66,67,68,69.

A CME signature was reported from a blueshifted emission component of the cool X-ray O viii line (4 MK) in the late phase of a stellar flare on the evolved giant star HR 902423. Although the time evolution of the blueshifted velocity is not obtained there, they detected the blueshifted emission component with a velocity of 90 km s−1 (escape velocity 220 km s−1) and interpreted it as a CME. The blueshifted plasma components at a few MK are also emitted from the upward flow in the confined flare loops (called ‘chromospheric evaporation’) in the case of solar flares, but they exclude the possibility considering that the other hotter lines do not show the blueshifted component in the post-flare phase. Although the spectral type of HD 9024 (evolved giant star) is very different from that of EK Dra and the velocity (90 km s−1) is smaller than our observation (510 km s−1), the two observations share the trend that the mass ejection signature is dominant in the post-flare phase.

Blueshifted emission components of chromospheric lines have been reported in association with Balmer-line flares mostly on active M/K dwarfs18,19,20,21,48,49,50,51,52,53,54,55,56,57 (see refs. 24,44 for a summary). Time-varying blueshifted hydrogen emission components have also been reported with high time cadence on M dwarfs (for example, refs. 19,21). A similar case is reported for a UV flare on an M dwarf20,70. These may be evidence of stellar prominence eruptions/CMEs. It seems quite possible that the blueshifted emission lines on M dwarfs are closely analogous to the Hα absorption signatures studied in this Letter. The fundamental difference between G-dwarf and M-dwarf blueshift signatures is that for hotter G dwarfs Hα in an erupting filament will only be detectable in absorption, whereas for the cooler M dwarfs even the quiescent Hα line is in emission, so an erupting filament might be observed in emission as well (compare ref. 44). Blue-wing enhancements of M-dwarf flares are characterized by high velocity of several hundred kilometres per second (sometimes more than this)18,53,55, which cannot be explained by chromospheric evaporation flow associated with the chromospheric-line blueshift phenomenon observed in solar flares16,44,71,72,73,74. The high velocities of M-dwarf flares are similar to that detected on EK Dra in this study (~510 km s−1). In addition, not all but some of the blueshift events on M dwarfs appear after the impulsive phase20,21, which shares properties with filament eruption events on EK Dra and the Sun in this study. Therefore, at present the blueshifted emission lines in M-type stars are most probably prominence eruptions.

Other signatures of kinematic characteristics of the ejected plasma are also inferred from continuous X-ray absorption during stellar flares, which can be caused by neutral material above the flaring region, such as filament eruptions24,59,60,75,76,77,78,79,80,81,82,83. However, on the Sun, X-ray absorption by prominences is uncommon44,84, and instrumental calibration effects at low energy have been pointed out47.

In some cases of binary stars, eclipses of the white dwarf component have been interpreted as obscuration by stellar mass ejected from the late-type companion star61,85. Other than this, pre-flare dips have been reported in stellar flares, suggesting potential prominence eruptions/CMEs86,87. Radio observations have recently investigated the type II radio bursts associated with shocks in front of CMEs as possible indirect evidence of CMEs, but no significant signature has been obtained so far47,62,63,64,65,67,68,69. Recently, a stellar type IV burst event from the M-type star Proxima Centauri was reported and may be evidence for a stellar CME56.