Complete replacement of magnetic flux in a flux rope
during a coronal mass ejection

The eruption of interest occurs in the NOAA active region 12158 on 2014 September 10 near the solar disk center. The active region consists of a leading positive-polarity sunspot surrounded by negative polarities in the southeast, and it exhibits a typical sigmoidal configuration (Fig. 1; Supplementary Video 1). A bundle of inverse-S shaped loops is observed several hours before the eruption in the SDO/AIA (Methods) 94 Å channel (Fig. 1d; Supplementary Video 2). The eruption starts with the rise of a coherent sigmoidal structure as observed in AIA’s hot passbands (e.g., 131 Å; Fig. 1b; Supplementary Video 1), which is interpreted as an MFR in previous studies on the event ^{7, 30, 31}. It produces a GOES X1.6-class (Methods) flare starting at 17:21 UT, and a fast halo coronal mass ejection (CME), which is detected as an ICME near the Earth two days after (Supplementary Note 1.3).

A sigmoidal MFR builds up from a thin thread in the core of a sheared magnetic arcade during the flare precursor phase (Fig. 1a,b), accompanied by footpoint brightening in the low solar atmosphere. The western foot of the pre-eruption MFR is located in positive polarities near the sunspot (Fig. 1g); the eastern foot in negative polarities is outlined by a low-atmosphere ribbon-like brightening (Fig. 1b,e). During the eruption, the same ribbon undergoes rapid and complex changes (Fig. 1c,f–h). Below we study in detail the formation of the sigmoid and its complex evolution during the eruption.

The eruptive sigmoid originally appears as a brightening inverse-S shaped thread in SDO/AIA 131 Å at $\sim$16:45 UT (Figs. 2&3), when the GOES 1–8 Å flux starts to rise slowly (Fig. 4a). The brightening thread is co-spatial with a low-lying filament (Fig. 2d), which stays almost unperturbed during the MFR eruption (Extended Data Fig. 1; Supplementary Note 1.2). The filament is aligned along the main polarity inversion line (PIL) of the active region, where a number of bald-patch (BP) candidates are identified (Fig. 2f; Methods). Embedded in a bundle of sheared sigmoidal loops observed in AIA 94 Å (Fig. 2c) and almost invisible in other cool AIA channels such as 211, 193 and 171 Å, the S-shaped thread has a mean temperature of about 7 MK with a dominant hot emission measure (EM) component at $\log T\approx$ 6.7–7.0, according to the differential emission measure (DEM) distribution sampled on the thread (labeled ‘S’ in Fig. 3; Methods). The EM loci curves of six extreme-ultraviolet (EUV) channels from the thread form an envelope over the multi-peak DEM distribution (Fig. 3b; Methods), corroborating the DEM result. The reference DEM distribution sampled from a nearby region off the thread (labeled ‘R’ in Fig. 3) shows similar cool EM components at $\log T\approx$ 5.5–6.6 but much weaker EMs at $\log T\approx$ 6.7–7.0, confirming the presence of hot plasmas in the thread. The newly appeared hot thread is later observed to continually grow into the eruptive sigmoid, hence is termed a ‘seed’ MFR hereafter ^{9, 12}.

As soon as it appears, the seed MFR further extends its “elbows” of the inverse-S shape and grows thicker (Extended Data Fig. 2; Supplementary Video 3) from several Mm to $\sim\,$30 Mm within the next $\sim$35 min before the eruption, as can be seen in the stack plot made from the virtual slit S1 across the sigmoid (Fig. 4c & Extended Data Fig. 2a; Methods). The growth of the seed MFR is associated with a slow rise in the GOES 1–8 Å SXR flux (Fig. 4a), before the onset of intense energy release at $\sim$17:21 UT. DEM analysis evidences intense heating during the formation of the MFR (Extended Data Fig. 3; Supplementary Note 1.1).

With the development of the MFR, significant pre-flare ribbon brightenings in the low atmosphere are observed by both SDO/AIA and IRIS (Extended Data Fig. 2; Supplementary Video 3). As the seed MFR extends to form a complete inverse-S shape, a trapezoid-shaped ribbon is spawned from the far end of the eastern ribbon (Extended Data Fig. 2). The spawning process is demonstrated in the stack plots of the virtual slits S2 and S3 (Fig. 4d,e & Extended Data Fig. 2d; Methods). The trapezoidal ribbon shows a fast and then slow expansion, which is associated with the fast and then slow widening of the coherent sigmoid observed in AIA 131Å (Fig. 4c–e). Comparing AIA 131 Å and IRIS 1400 Å images at about 17:12 UT (Fig. 5a,d & Extended Data Fig. 2), one can see that the trapezoidal ribbon is the footpoints of coronal loops belonging to the sigmoidal MFR. That the expansion of the trapezoidal ribbon is accompanied by the sigmoid thickening argues strongly for the development of a twisted MFR ^{2, 3}. During this phase, the instantaneous slipping motion proceeds almost perpendicularly to the overall direction of the ribbon extension, which is described as “squirming” motions in a previous study ³⁰. During the flare precursor phase, the rate of the magnetic reconnection that contributes to the MFR buildup is estimated to vary around $2\times 10^{19}~{}\mathrm{Mx}~{}\mathrm{s}^{-1}$ (Fig. 4b; Methods), a few times lower than, but still on the same magnitude as, that during the MFR eruption.

Shortly after being well developed, the MFR erupts to produce an intense X-class flare. The flare ribbon in the positive polarity region (PR) remains relatively stationary (Fig. 1g, Extended Data Fig. 4b), probably due to its proximity to the sunspot where the strong magnetic field is relatively “rigid”; the ribbon in the negative polarity region (NR), which is associated with the eastern foot of the MFR, evolves greatly and rapidly during the eruption (Fig. 1g,h & Extended Data Fig. 4; Supplementary Video 3).

From about 17:18 UT, NR’s hook changes from a closed to an open morphology (Fig. 5, Extended Data Fig. 4; Supplementary Video 3). As the ribbon moves, it sweeps the trapezoidal ribbon that encloses the eastern foot of the pre-eruption MFR. During the eruption, both ends of NR elongate rapidly southwestwards, and the whole ribbon develops into an horseshoe shape (Extended Data Fig. 4). The northern branch of NR (NR_n), which host negative footpoints of post-flare loops, extends rapidly along the direction of the main PIL; the motion is associated with the strong-to-weak shear transition of the post-flare arcade observed in AIA (Extended Data Fig. 4; Supplementary Video 3). Sometimes NR_n itself shows two parallel branches in IRIS 1400 Å (e.g., at 17:33 UT; Extended Data Fig. 4d), but they are not resolved in AIA 1600 Å observations. The southern part of the horseshoe-shaped ribbon (NR_s) completes the hooked part of NR, which supposedly maps the footprint of the QSLs that wrap around the erupting MFR ²⁰.

In particular, NR_s exhibits irregular, complex motions (Fig. 1h; Supplementary Video 3). It zigzags towards southwest, in a serpentine fashion. The stack plot of the virtual slit S4 across the horseshoe-shaped ribbon (Extended Data Fig. 4d) nicely shows the zigzag motions of NR_s, which leave irregular trails, distinct from the rather smooth track left by NR_n (Fig. 4f). As the eruption proceeds, the horseshoe shape is stretched, but its opening becomes wider, where coronal dimmings develop (Fig. 5, Extended Data Fig. 4). During the flare decay phase, NR_s fades away, and only NR_n remains visible in AIA 1600 Å images (Figs. 1g&4f). During the eruption, some bright knots are observed to distribute quasi-periodically and move along the ribbon, which may indicate that the slipping reconnection proceeds quasi-periodically ³³.

The ribbon brightening in IRIS SJI 1400 Å images shows three phases of temporal evolution (Fig. 6), i.e., 1) the development of the trapezoidal ribbon from 16:45 to 17:18 UT, 2) the change of NR’s hook from a closed to an open morphology from 17:18 to 17:26 UT, and 3) the evolution of the horseshoe-shaped NR afterwards. It is remarkable that most of the region enclosed by the trapezoidal ribbon (marked by orange dotted lines in Fig. 6) that is swept during the first phase is swept again during the second phase (as shown in blue in Fig. 6d), which suggests that the reconnections taking place during the second phase must have involved most of the magnetic flux in the MFR formed during the first phase. It is also remarkable that NR_n, the ribbon segment that hosts the negative footpoints of flare loops, only appears during the third phase (Fig. 6c,d). In contrast, in a different event featuring an MFR under formation during the eruption ³, the two straight flare ribbons appear first and later extend and develop a closed hook at the far end of each ribbon, enclosing the rope’s two feet.

Enclosed by flare ribbons, two dimming regions are observed in cool AIA EUV passbands (Extended Data Fig. 5), indicative of coronal mass loss along two feet of the CME. The dimming in negative polarities as enclosed by NR is evident in AIA 335 Å images, and it migrates as the MFR erupts (Fig. 5b,c; Extended Data Fig. 4c; Supplementary Video 4). The dimming lightcurve (Fig. 4a; Methods) shows that coronal dimming around NR starts to develop from $\sim$17:10 UT onward, when the MFR is well developed as characterized by its eastern foot being fully outlined by a trapezoidal ribbon. However, the initial dimming is not enclosed by, but located outside, the trapezoid (Fig. 6e; Methods). During the eruption, in spite of strong diffraction patterns in AIA images, the dimming intensifies as shown by the 335 Å lightcurve (Fig. 4a). The coronal dimming mainly occupies within the horseshoe-shaped ribbon, close to the NR_s side (Fig. 6f,g). As the horseshoe shape of NR is stretched, the dimming region is mainly concentrated in the opening of the horseshoe after 17:50 UT (Fig. 6h), and it stays quasi-stationary thereafter during the long-duration decay phase (Supplementary Video 4). A conjugated pair of dimmings are identified on both the northern and southern sides of the post-flare arcade during the late decay phase (Extended Data Fig. 6). Corresponding to the extended decreasing brightness (Fig. 4a; Methods), these dimmings continue to map the two footpoints of the CME, whose connection settles in the north-south direction (Extended Data Fig. 6d), in contrast to the orientation of the pre-eruption MFR that is mainly along the east-west direction (Extended Data Fig. 6a).

Combining the evolution of flare ribbon and coronal dimming (Fig. 6), we conclude that the eastern foot of the MFR undergoes a drastic migration during the eruption. The centroid of magnetic fluxes through the foot is shifted by about 70^′′ during the impulsive eruption (Fig. 6h; Methods). Its average migrating speed is about 26 km s^-1, significantly faster than the typical Alfvén speed in the lower solar atmosphere ($\sim$10 km s^-1) and the typical speed of photospheric flows ($\sim$1 km s^-1). Considering the irregular and highly dynamic motions of the ribbon hook (Fig. 1h; Supplementary Video 3), the instantaneous migrating speed would also be irregular and most likely bursty. The magnetic flux through the foot, which gives an estimation of the toroidal flux of the rope, increases from about $3.3\times 10^{20}$ Mx to $7.9\times 10^{20}$ Mx during the flare impulsive phase (Methods). Thus, the substantial displacement of and the associated twofold increase in magnetic flux through the foot, strongly evidence a complete restructuring of the original MFR during its development into the CME. This is radically different from the continuous deformation and drifting of flux-rope footpoints found in previous studies ^{34, 35}.

The Atmospheric Imaging Assembly ⁴³ (AIA) onboard the Solar Dynamics Observatory ⁴⁴ (SDO) takes full-disk images of the Sun around the clock. The seven EUV channels, i.e., 131 Å (primarily from Fe XXI emission line, $\log T=7.05$), 94 Å (Fe XVIII, $\log T=6.85$), 335 Å (Fe XVI, $\log T=6.45$), 193 Å (primarily Fe XII, $\log T=6.2$), 211 Å (Fe XIV, $\log T=6.3$), 171 Å (Fe IX, $\log T=5.85$), and 304 Å (He II, $\log T=4.7$), obtain images with a spatial resolution of $0.^{\prime\prime}6$ and a temporal cadence of 12 s. Two UV channels, 1600 Å (C IV line and continuum emission, $\log T=5.0$) and 1700 Å (continuum), obtain images with a temporal cadence of 24 s. AIA data are processed to level 1.5, and the heliographic maps are all rotated to 16:00 UT to correct the solar rotation.

The Helioseismic and Magnetic Imager ⁴⁵ (HMI) onboard SDO measures the magnetic fields on the solar photosphere. We use HMI line of sight magnetograms to compare with heliographic observations by AIA and IRIS. To investigate the detailed magnetic configuration of the active region, we use the HMI active region patches (SHARP) vector magnetograms, which are taken every 12 minutes and remapped with the cylindrical equal area (CEA) projection, with a pixel scale of about 0.36 Mm. At the PIL of the active region which is identified by the contour of $B_{z}=0$, we search for the existence of bald patches (BPs; Fig. 2f) by employing the criterion $(B_{\bot}\cdot\nabla_{\bot}B_{z})|_{\rm{PIL}}>0$ ^{46, 47}, where magnetic field vectors are directed from negative to positive polarities and the field line is tangential to the photosphere and concave upward. BPs are topological structures known to be favorable for the formation of current sheets and therefore for the occurrence of magnetic reconnection (see also Supplementary Note 1.2).

The Interface Region Imaging Spectrograph (IRIS)⁴⁸ captures the formation and eruption phase of the event before 17:58 UT. The Slit Jaw Imager (SJI) obtains images with a 19-s cadence and a spatial resolution as high as $\tfrac{1}{6}^{\prime\prime}$ over a field of view (FOV) of $119^{\prime\prime}\times 119^{\prime\prime}$, which fortuitously covers the eastern foot of the MFR (Supplementary Video 3). IRIS level-2 data of SJI Si IV 1400 Å passband ($\log T\approx$ 4.9, forming in the transition region) are used in the study. The superior spatiotemporal resolution of IRIS is key to this study; e.g., it registers brightening in 1400 Å around the eastern footpoint of the seed MFR as soon as it appears in EUV at 16:45 UT (Fig. 2e), suggestive of simultaneous heating in the low atmosphere. However, the brightening is too thin to be clearly resolved by AIA ultraviolet (UV) channels during this early phase. The subsequent footpoint evolution of the seed MFR is resolved by both IRIS and SDO/AIA, but the morphology and evolution of flare ribbons are better resolved by IRIS both temporally and spatially. IRIS observations hence serve as a reference and guide for our identification of flare ribbons observed by AIA. Together, IRIS and AIA observations make it possible to estimate the pre-flare reconnection rate (see below), which is rarely given in the literature.

To characterize the temporal evolution, we place four virtual slits (S1–S4; Extended Data Figs. 2&4) in SDO/AIA and IRIS images and generate the time–distance stack plots (Fig. 4c–f). S1 is put across the sigmoid in SDO/AIA 131 Å images (Extended Data Fig. 2a), with a length of 76 Mm, averaged over 6 AIA pixels. S2 and S3 are across the trapezoid-shaped ribbon in IRIS 1400 Å (Extended Data Fig. 2d), with a length of 47 and 59 Mm, respectively, averaged over 10 IRIS pixels. S4 is across the horseshoe-shaped ribbon in IRIS 1400 Å (Extended Data Fig. 4d), with a length of 45 Mm, averaged over 10 pixels. Stack plots of S2–S4 after 17:58 UT are generated from SDO/AIA 1600 Å images at the same locations, when there are no IRIS observations.

The Geostationary Operational Environmental Satellite (GOES) spacecraft operates in geosynchronous orbit and measures the solar soft X-ray flux in units of $\text{W}\,\text{m}^{-2}$. The classification of solar flares uses the letters A, B, C, M, or X according to the peak flux in 1–8 Å as measured by GOES, with each letter representing a 10-fold increase in flux.

SDO/AIA’s six optically thin EUV passbands (except for 304 Å) are used for differential emission measure (DEM) analysis to study the temperature characteristics. The AIA data are further processed to level 1.6 before being fed into the DEM code. Here we used the sparse inversion code ⁴⁹, which has been further modified to optimize solutions for hot plasmas above a few MK⁵⁰. DEMs are calculated in the temperature range of $\log T$ = 5.5–7.5 with an interval of $\Delta\log T$ = 0.05, and a DEM-weighted mean temperature, $\langle T\rangle=(\sum DEM(T)\times T\Delta T)/(\sum DEM(T)\Delta T)$, is also derived ⁵¹. DEM uncertainties are obtained by 250 times of Monte Carlo simulations. We also computed the EM loci curves ($\mathrm{EM}_{\lambda}(T)=I_{\text{obs}}/\epsilon_{\lambda}(T)$) of a sampled region using the observed intensities $I_{\text{obs}}$ and temperature response functions $\epsilon_{\lambda}(T)$ from the six AIA channels.

To characterize the ribbon evolution, we identify the low-atmosphere ribbon by counting brightened pixels in SDO/AIA UV channels (Fig. 1g) and IRIS SJI 1400 Å passbands (Fig. 6), respectively. During the precursor phase, flare brightening in AIA 1600 Å is too weak to be clearly distinguished from chromospheric plages, thus we use the ratio of AIA 1600 Å and 1700 Å passbands to enhance the ribbon brightening ³⁰ (Extended Data Fig. 2; Supplementary Video 3). After the flare onset at 17:21 UT, only AIA 1600 Å passband is used. To best identify the flare ribbons, different threshold values are set during different evolutionary phases.

The identified ribbon pixels in SDO/AIA are used to calculate the reconnection flux (Fig. 4b), after being projected onto a pre-flare SDO/HMI SHARP $B_{z}$ map with the CEA projection. We measured the instantaneous reconnection flux by summing up magnetic fluxes swept by newly brightened ribbon pixels, and estimated the uncertainties by the difference of the measured magnetic flux in positive and negative polarities. Integrating the instantaneous reconnection flux with respect to time, we obtained the accumulative reconnection flux.

Compared to AIA, IRIS is superior in spatial resolution and dynamic range, though it has a limited field of view mainly covering the ribbon of negative polarity (Extended Data Figs. 2 & 4, Supplementary Video 3). We manually traced the bright ribbon fronts observed in IRIS 1400 Å at selected time instants to highlight the complex patterns and irregular motions (Fig. 1h).

Coronal dimming around the ribbon NR are evident and migrate in SDO/AIA 335 Å passband during the MFR eruption (Extended Data Fig. 4; Supplementary Video 4). To study the dimming evolution, we identified dimmings as the pixels whose brightness decreases compared to that before the eruption (we refer to 16:20 UT for the event; Fig. 6e–h), and took the average brightness to obtain the 335 Å dimming lightcurve (Fig. 4a). For the conjugated pair of post-eruption dimmings detected at the footpoints of large-scale overlying loops (Extended Data Fig. 6d), we sampled intensities at two representative locations in SDO/AIA 171 Å (‘DS’ and ‘DN’ in Extended Data Fig. 6d, $10^{\prime\prime}\times 10^{\prime\prime}$) to show their temporal evolution (Fig. 4a).

To quantitatively study the footpoint migration, we compare two time instants before and after the eruption as an illustration, when the eastern foot of the MFR is well defined; i.e., at about 17:18 UT when it is enclosed by a trapezoidal ribbon (Fig. 6a), and after 17:50 UT when it is marked by the coronal dimming that is concentrated in the opening of the ribbon hook (e.g., Fig. 6h). One can see that the centroid of magnetic fluxes within the foot migrates from about [-167^′′, 111^′′] to [-108^′′, 76^′′] (Fig. 6h), which is shifted by about 70^′′ with an average migrating speed of about 26 km s^-1. The magnetic flux through the foot changes from about $3.3\times 10^{20}$ Mx to $7.9\times 10^{20}$ Mx accordingly, which increases nearly twofold. While the foot boundary is difficult to determine at other times during the eruption, the substantial displacement of the foot and the increase of magnetic flux through it strongly evidence a complete restructuring of the original MFR during its development into the CME.

The data used in the study are publicly available for download from the corresponding mission archives. SDO data are available at http://jsoc.stanford.edu/ajax/lookdata.html; IRIS data are available at https://iris.lmsal.com/data.html.

We acknowledge SDO and IRIS teams for the science data. IRIS is a NASA small explorer mission developed and operated by LMSAL with mission operations executed at NASA Ames Research Center and major contributions to downlink communications funded by ESA and the Norwegian Space Centre. T.G. thanks the BBSO staff for providing H$\alpha$ data. T.G., R.L., W.W., M.X., and Y.M. acknowledge the support from National Natural Science Foundation of China (NSFC 42188101, 42274204, 11903032, 11925302, 12003032) and the Strategic Priority Program of the Chinese Academy of Sciences (XDB41030100). A.M.V. acknowledges the Austria Science Fund (FWF): project I-4555N. T.L. acknowledges the support from NSFC 12222306. T.G. also acknowledges the support from CAS Key Laboratory of Solar Activity.

T.G. and R.L. led the study and analysis, interpreted the data, and wrote the manuscript. A.M.V. discussed the analysis and contributed to the interpretation, conclusion and writing of the manuscript. B.Z. led the in-situ data analysis and discussed the interpretation. T.L. and Y.W. discussed the results and contributed to the interpretation. W.W. and M.X. contributed to the SDO and in-situ data analyses. All authors participated in the discussion and contributed to finalizing the manuscript.

The authors declare no competing interests.

We observe in this event that a coherent sigmoidal flux rope builds up upon a slim S-shaped seed, which appears beneath a sheared arcade in the active region, accompanied by simultaneous ribbon-like brightenings in the low solar atmosphere (Fig. 2). The flux rope grows continually from the hot S-shaped seed within 30–40 minutes during the flare precursor phase (Extended Data Fig. 2). The buildup process manifests itself in the continual lengthening and thickening of the S-shaped loop in the AIA 131Å passband, and in its eastern footpoint slipping away from the core field and subsequently expanding into a trapezoidal ribbon in IRIS SJI 1400Å (Extended Data Fig. 2; Supplementary Video 3), which argues strongly for the development of a twisted magnetic flux rope ^{2, 3}. While tether-cutting reconnections ^{4, 5} may help produce the S-shaped thread, as evidenced by the persistent flux cancellation ⁶ in this decayed active region ⁷, the subsequent growth of the sigmoid is mainly featured by the footpoint slippage (Extended Data Fig. 2; Supplementary Video 3), which suggests that 3D slipping reconnections ⁸ play an important role.

We studied the temperature characteristics of the sigmoid and its evolution via DEM analysis. The results (Extended Data Fig. 3; Methods) show that the flux rope structure has a mean temperature as high as 10 MK. The flux-rope plasma appears to consist of a hot core at 10–20 MK and a less hot envelope at 5–10 MK, subject to the confusion of the line-of-sight integration through the optically-thin corona. Before the seed flux rope appears (the left column in Extended Data Fig. 3), inverse-S shaped loops observed in AIA 94 Å are evident in the 5–10 MK temperature range. The DEM sampled from the sigmoidal flux rope (labeled as ‘S’ in Extended Data Fig. 3) shows that the hot DEM component with $\log T\approx$ 6.8–7.2 increases greatly in magnitude with time, while the cooler components with $\log T<$ 6.8 decrease. In comparison, the DEM sampled from a nearby location off the rope for reference (labeled as ‘R’) changes little with time. Thus, intense heating must be involved in the formation of the flux rope.

That the S-shaped thread continually grows into a bulky sigmoid and later into a CME (Supplementary Video 1) is reminiscent of the flux-rope buildup process reported in Gou et al. (2019) ⁹, in which a plasmoid at the upper tip of a pre-existent current sheet embedded in a sheared magnetic arcade continually grows into an oval-shaped bubble and later into a CME. Complementing the side view of the limb eruption on 2013 May 13 in Gou et al. (2019)⁹, the present on-disk sigmoidal eruption provides a top view of the flux rope formation process. The width of the seed thread is about 2–3^′′ (Fig. 2), similar to the scale of the leading plasmoid in Gou et al. (2019)⁹. The temperature structures of the seeds in the two events are also similar, which contain significant hot emissions in the temperature range of $\log T\approx$ 6.7–7.2 (Fig. 3b and Fig.2 in Gou et al. 2019⁹). However, the mean temperature of the flux rope in the present event (Extended Data Fig. 3c) could be underestimated due to the contribution from a plethora of overlying loops above it. In addition, owing to the on-disk observation and the unprecedented resolution of IRIS, we observe lower-atmosphere brightenings that evolve synchronously with the development of the sigmoidal flux rope (Extended Data Fig. 2). Such brightenings map the footpoints of coronal magnetic field lines undergoing magnetic reconnection, but are seldom observed during the flare precursor phase ¹⁰, therefore providing valuable information on the flux-rope buildup process prior to the eruption, which complements the flux-rope buildup process during the eruption ³.

In contrast to two competing pre-eruption magnetic field configurations of CMEs, i.e., sheared arcade vs. flux rope ¹¹, the observations that a large-scale flux rope initiates from a seed in both on-disk (e.g., the present study and Wang et al. 2017 ³) and limb (e.g., Gou et al. 2019⁹) events suggest a hybrid state with a tiny twisted core embedded in a sheared arcade ¹². These observations confirm the speculation that there could exist a continuum of intermediate states between the two distinct magnetic configurations ¹¹; however, it remains unclear how such an arcade-to-rope transition exactly occurs. The appearance of the seed in the hybrid scenario represents a sudden commencement of the transition from a sheared magentic arcade to a coherent flux rope that starts to build up upon the seed. The buildup process can occur either before (e.g., the event under study) or during the eruption ^{3, 9}. The seed may be a key structure for the the production of highly twisted flux ropes ¹³ that are detected on the Sun ³ or in interplanetary space ¹⁴.

The central part of the pre-eruption flux rope is aligned with the main PIL of the active region, which is co-spatial with a filament observed in the H$\alpha$ line center by the Big Bear Solar Observatory (BBSO; Fig. 2d and Extended Data Fig. 1). The filament remains intact beneath the post-flare arcade (Extended Data Fig. 1), which indicates a partial eruption of a double-decker structure in the active region. The lower branch, i.e., the filament, could be either supported by a sheared magnetic arcade or embedded in a flux rope ¹⁵. We find a number of candidate BP points at the photospheric PIL that is occupied by the filament (Fig. 2f). In addition, significant magnetic flux cancellation occurs around the PIL long before the eruption ⁷, which may produce the BP configuration of magnetic fields ⁶. Thus, the BPs may help hold down the filament ¹⁶. Both the dextral filament and the inverse S-shaped sigmoid are likely associated with negative magnetic helicity ^{17, 18, 19}, which also agrees with the accumulation of negative helicity in the active region before the eruption ²⁰.

A question arises as to how and when the double-decker configuration forms. The two branches of the double decker could be adhered to each other in equilibrium before the eruption but split near or during the eruption ²¹. In this event, we observed that the upper branch of the erupting flux rope originally appears as a shorter thread apparently superposing the low-lying filament (Fig. 2); its growth associated with footpoint slipping is characterized by the increase of SXR flux and footpoint brightening of chromospheric heating (Fig. 4 & Extended Data Fig. 2). The footpoint slippage suggests that 3D magnetic reconnection is responsible for the formation of the upper flux rope during the precursor phase before the eruption. Cheng et al. (2015) ⁷ suggests magnetic reconnection occurs in the low solar atmosphere based on the observed Doppler blue shifts. Such reconnections at the bald-patch separatrix surface (BPSS), which consists of magnetic field lines touching the BP points at the photosphere, may lead to the formation of a flux rope atop the structure associated with the filament ^{16, 22}.

The fast full halo CME produced by the sigmoid eruption arrives at the Earth two days after. The Advanced Composition Explorer (ACE) spacecraft observes an interplanetary coronal mass ejection (ICME) structure between Sep 12 22:30 UT and Sep 14 12:00 UT, preceded by a shock front driven by the ICME at 15:30 UT on Sep 12 (Supplementary Fig. 1). In-situ observations show ICME features ²³ such as enhanced magnetic fields, decreases in proton number density, temperature and plasma $\beta$ (Supplementary Fig. 1). The bi-directional distribution of suprathermal electrons (Supplementary Fig. 1h) indicates that the interplanetary structure remains being magnetically connected to the Sun at both ends.

Some features of the ICME also conform to the MC criteria, and it is identified as an MC in a few ICME catalogs ^{24, 25, 26}. However, it does not possess a typical flux-rope configuration. For instance, the magnetic field vectors do not exhibit a large ($\gg 30^{\circ}$) and smooth rotation as typically in MCs ^{27, 28} (Supplementary Fig. 1b & c). Instead, the in-situ observation reveals multiple discontinuities with jumps in both magnetic field and plasma parameters inside the ICME (marked by vertical dotted lines in Supplementary Fig. 1). Some flux-rope models were used to reconstruct the global magnetic configuration of the ICME. A circular-cylindrical analytical flux-rope model fitting, which was included in a statistical study, obtains a magnetic obstacle with an incomplete rotation less than 90^{∘ 25}. A torus-shaped flux rope model fitting indicates that the spacecraft may skim the ICME flux rope, covering so small a field rotation that the ICME can be fitted under the assumption of either a left-handed or a right-handed helicity ^{29, 30}. We also modeled the ICME with two state-of-the-art techniques but did not obtained reasonable results. We first used the velocity-modified cylindrical flux-rope models, adopting either a non-uniform (linear force-free model with Lundquist solution) ³¹ or a uniform (non-linear force-free model) ¹⁴ twisted magnetic field configuration. Both fittings match poorly with the observations (the normalized root-mean-square difference between the fitting results and observations $\chi_{n}>0.5$), especially for the magnetic field magnitudes and directions. We also used the Grad-Shafranov (GS) reconstruction technique ³², which does not presume any specific shape of the ICME cross section, but still failed to reconstruct a magnetic flux rope structure. The unsuccessful fittings might be due to limitations of the models, flank crossing of the spacecraft, or complex structures of the ICME itself.

In this event, the complete replacement of magnetic flux in the preformed flux rope in a drastic fashion makes it hard to imagine that the flux rope can preserve a well-organized structure during its development into the (I)CME. Indeed, the ICME does not possess a typical flux-rope configuration but multiple internal discontinuities (Supplementary Fig. 1). In-situ detected complex ejecta have been attributed to unfavorable spacecraft trajectories ³³, interactions with other ejecta ³⁴ or solar wind ³⁵, and the complexity of pre-eruptive structures ³⁶. Here we suggest that the complex flare reconnections that reshape the flux rope can also contribute to the complexity of the resultant ICME, which agrees with numerical experiments that produce an inherently complex CME ³⁷ and may help understand why only about one third of ICMEs possess flux-rope configurations.