Intrusion of Magnetic Peninsula toward the Neighboring Opposite-polarity Region That Triggers the Largest Solar Flare in Solar Cycle 24

The X2.2 and X9.3 flares that are the focus of this study commenced at 08:57 and 11:53 UT, and peaked at 09:10 and 12:02 UT, respectively, on September 6, 2017. Hinode (Kosugi et al., 2007) had tracked the AR constantly from approximately 11:00 UT on September 5 until it disappeared from sight on September 14, and perfectly observed the X2.2 and X9.3 flares on September 6 using its three instruments: Solar Optical Telescope (SOT, Tsuneta et al., 2008), Extreme-ultraviolet Imaging Spectrometer (EIS, Culhane et al., 2007), and X-ray Telescope (XRT, Golub et al., 2007).

Hinode/SOT determined the full polarization states (Stokes-I, Q, U, and V) at 6301.5 Å and 6302.5 Å (Fe I line) with a sampling of 21.5 mÅ using a Spectro-Polarimeter (SP, Lites et al., 2013). We used an SP map, which was scanned from 06:14 UT on September 6, 2017, to analyze three components of the photospheric magnetic field prior to the onset of the two X-class flares. The SP map was obtained with the fast-mapping mode observation that covered a field-of-view of 164${}^{\prime\prime}\times$ 164^′′. The sampling was 0.3^′′ along both the slit and slit-scan.

Hinode/EIS successfully observed both the X2.2 and X9.3 flares. We investigated the local Doppler velocities of plasma related to the precursor brightening using the data from EIS. The raster scan was repeated from 06:20 to 15:35 UT on September 6, 2017, for 10 different wavelengths. In this study, we mainly used Fe XIV lines corresponding to 264.79 Å and 274.20 Å, Fe XVI/Fe XXIII lines corresponding to 263.3 Å, and He II lines corresponding to 256.3 Å. The Fe and He II lines are sensitive to emissions from $log(T)\sim 6.3-6.4$ plasmas and $log(T)\sim 4.9$ plasmas, respectively (Culhane et al., 2007). The $2^{\prime\prime}\times 120^{\prime\prime}$ slit scanned 30 positions with sparse $4^{\prime\prime}$ steps, where $4^{\prime\prime}$ is the distance between the center of the slits; hence, the scanning field-of-view covered $120^{\prime\prime}\times 120^{\prime\prime}$. The exposure time at each position was 4s, and the spectral resolution was 22 mÅ.

Hinode/XRT continuously observed the AR using a standard AR observation program; this program captures coronal images in multiple wavelengths with high time cadence. We investigated the high-temperature coronal loops that appeared to relate to the magnetic reconnection among sheared magnetic fields using the XRT images that were captures with a thin-Be/Al-poly filter covering a $384^{\prime\prime}\times 384^{\prime\prime}$ field-of-view. The spatial resolution was $2^{\prime\prime}$, and the data cadence of each thin-Be/Al-poly images was approximately 90s.

The Solar Dynamics Observatory (SDO, Pesnell et al., 2012) observed the X2.2 and X9.3 flares using the Helioseismic and Magnetic Imager (HMI, Schou et al., 2012) and the Atmospheric Imaging Assembly (AIA, Lemen et al., 2012). We used the SDO data to investigate the temporal evolution of the photospheric magnetic fields and the precursor activities in the solar atmosphere. Moreover, we primarily used vector magnetic field data referred to as the Space-weather HMI AR Patch (SHARP¹¹1“hmi.sharp_720s” series downloaded from Joint Science Operations Center (JSOC) at Stanford University (http://jsoc.stanford.edu/), Bobra et al., 2014) data series. The SHARP data series has been calibrated assuming the Milne-Eddington atmosphere; its $180^{\circ}$ ambiguity in the horizontal component was resolved via the minimum energy method (Metcalf, 1994; Leka et al., 2009b, 2012). The SHARP data was continually obtained every 12 min, and we considered the data obtained from 00:00 UT on September 6, to 00:00 UT on September 7, 2017, for use in this study. We also used AIA level 1 data²²2“aia.lev1_euv_12s” and “aia.lev1_uv_24s” series that was obtained in the time window of 00:00 UT on September 6, to 00:00 UT on September 7, 2017. AIA observed the AR with 10 wavelengths, but we mainly analyzed the images captured corresponding to 1600Å and 171Å. We used AIA 211Å and 335 Å data to spatially co-align the HMI/SHARP magnetic field data and Hinode/EIS or XRT data (see details in Section 2.2). The data cadence was 24s for 1600Å and 12s for 171Å, 211Å, and 335Å. In addition, we used HMI level 1.5 line-of-sight magnetic field data³³3“hmi.M_45s” series that was observed via HMI filtergram observation for the Fe I line (6173Å) with a 45s cadence in the same time window as that of AIA data to superpose SHARP vector magnetograms and AIA or Hinode/EIS or XRT images. The spatial resolution was $1.0^{\prime\prime}$ for SHARP and HMI/line-of-sight magnetograms, and $1.5^{\prime\prime}$ for AIA images. Note that we avoided using the SDO data obtained from 06:22:34 UT to 07:53:11 UT on September 6, 2017, because these data were not of good quality owing to the lunar transit.

The Hinode/SOT-SP data was calibrated via the sp_prep procedure (Lites, & Ichimoto, 2013) using the Solar-SoftWare (SSW) package (Freeland, & Handy, 1998). Subsequently, vector magnetic fields were derived from the calibrated Stokes profiles based on the assumption of the Milne-Eddington atmosphere using the inversion code MERLIN at HAO/CSAC (DOI: 10.5065/D6JH3J8D). The minimum energy code ME0 (Leka et al., 2009a, b, 2012), which is based on the minimum energy method of Metcalf (1994), was adopted to resolve the 180^∘ ambiguity in the tangential component of the magnetic fields. As highlighted by Leka et al. (2017); Bamba, & Kusano (2018), the three components of the photospheric magnetic field are influenced by a projection-effect; that is, the line-of-sight component of the magnetic field can be treated as the radial component only when the observing angle $\theta=0$. The AR 12673 was located on September 6, 2017, at S09W42, which is close to the south-east limb. Therefore, to investigate locations and shapes of magnetic polarity inversion lines, we transformed the vector magnetic field data obtained via SOT-SP, which is in image plane on the Charge Coupled Device (CCD), to its corresponding heliocentric coordinates. To transform the magnetic field vectors, we used the transformation matrix presented by equation (1) of Gary, & Hagyard (1990). In this manner, we obtained the radial component ($B_{z}$) and the horizontal components ($B_{x},B_{y}$) of the photospheric magnetic field vectors.

Hinode/EIS raster scan data was calibrated via the eis_prep procedure using the SSW package. For reasons concerning thermal factors, there exists an orbital variation of the line position causing an artificial Doppler shift of $\pm 20km/s$, which follows a sinusoidal behavior. This orbital variation and wavelength calibration of the line position were corrected using the house keeping data (Kamio et al., 2010). Subsequently, we first performed wavelength calibration using the Fe XIV (264.787Å) line, following which we finalized the central value $\lambda^{\prime}_{obs}$ for the observed Fe XIV line profiles via the single Gaussian fitting. Herein, we integrated the observed line profile over the region that was relatively quiet as per the EIS scanning field-of-view. We then applied the single Gaussian fitting to the integrated line profile and repeated the process for the data that were scanned from 06:21:03⁴⁴4 UT to 06:58:03⁴⁴4It means scan start time. UT on September 6, 2017. We averaged the value of fitted line center over the abovementioned time period and obtained the observed line center $\lambda_{obs\_FeXIV}$. Subsequently, we determined dw by subtracting $\lambda_{obs\_FeXIV}$ from $\lambda_{lit\_FeXIV}$, which is the literature wavelength of the Fe XIV line as per literature (Brown et al., 2008): $dw=\lambda_{lit\_FeXIV}-\lambda_{obs}$. Using dw, we calibrated the other spectral lines as per $\lambda_{0}=\lambda_{lit}-{\it dw}$, where $\lambda_{lit}$ is the wavelength of the other lines based on literature (Brown et al., 2008), and $\lambda_{0}$ is the reference wavelength of those lines as per the EIS raster scan observation. Following this, we identified the center of observed spectral lines $\lambda_{obs}$ via the eis_auto_fit procedure using the SSW package, which applies single Gaussian fitting to the line profiles. We then generated intensity, Doppler velocity, and line width maps for each of the observed lines. The Doppler velocity was calculated as per $v={\Delta\lambda}c/{\lambda_{0}}$, where $\Delta\lambda=\lambda_{0}-\lambda_{obs}$, and $c$ is the speed of light. Using the intensity, Doppler velocity, and line width maps, we checked the locations and timings of precursor brightening occurrences relative to the two X-class flares. We further applied double Gaussian fitting to the averaged line profile of the Fe XIV lines (264.7Å and 274.2Å) at the single-slit position of the precursor brightening and measured the Doppler velocity of a local and transient flow, which is described in Section 3.2. Note that the blend from the other lines is negligibly small for Fe XIV (264.7Å) and Fe XVI (262.9Å) (Young et al., 2007).

The Hinode/XRT data was calibrated via the xrt_prep procedure (Kobelski et al., 2014) using the SSW package. Subsequently, we first spatially co-aligned the XRT Al-poly images and SDO/AIA 335Å, which show structures similar to each other (Yoshimura & McKenzie, 2015). The AIA 335Å images have been spatially well aligned with the HMI/line-of-sight images via the aia_prep procedure, as we described in Section 2.3. Thus, we aligned the radial magnetic field ($B_{z}$) images of Hinode/SOT-SP or SDO/HMI with the line-of-sight magnetic field images. In this manner, we finally overplotted the XRT intensities onto the $B_{z}$ field images or vice versa (Figure 6).

HMI/line-of-sight magnetograms and AIA data were calibrated via the aia_prep procedure using SSW, whereby can be rotated such that the solar EW and NS axes are aligned with the horizontal and vertical axes of the image. The pixel scales, which are different between HMI and AIA, were resampled to the same size. Thus, the positions of the line-of-sight magnetograms and AIA images positions were aligned. Next, we transformed SHARP vector magnetograms to the heliocentric coordinates in the same manner for Hinode/SOT-SP. Therefore, we obtained the radial component ($B_{z}$) and horizontal components ($B_{x},B_{y}$) of the photospheric magnetic field vectors. We then extracted an line-of-sight magnetogram, SHARP vector magnetogram, and AIA image, which were the closest in terms of time, and superposed the SHARP vector magnetic field and AIA image via the line-of-sight magnetogram. Moreover, we superposed the magnetic polarity inversion lines of the radial component ($B_{z}$) of the SHARP data onto the Hinode/EIS maps corresponding to the intensity, Doppler velocity, and line width maps. For this superposition, the Hinode/EIS maps corresponding to the Fe XIV (264.787Å) line and AIA 211Å images were superposed. The AIA 211Å images and the SHARP magnetic field data had already been co-aligned via the line-of-sight magnetic field data; thus the magnetic polarity inversion lines of $B_{z}$ from the SHARP data were overplotted onto the Hinode/EIS maps. We showed the magnetic polarity inversion lines on the EIS maps, which were observed during the precursor brightening (Figure 4). Then magnetic polarity inversion lines of $B_{z}$ fields from the SHARP were also overplotted on the Hinode/XRT images (Figure 6) in the same manner, although herein we employed AIA 335Å instead of 211Å, as we mentioned in Section 2.2.

We extrapolated three-dimensional (3D) coronal magnetic fields based on the photospheric magnetic fields before the two X-class flares. We used the modeling of the 3D coronal magnetic field using an NLFFF approximation, as described in Wiegelmann & Sakurai (2012), and employed the MHD relaxation method developed in Inoue et al. (2014); Inoue (2016). The SHARP vector magnetic fields ($B_{x},B_{y},B_{z}$) observed at 08:36 and 11:00 UT on September 6, 2017, were set as the bottom boundary condition. We solved the MHD equations in 3D space starting with a potential field, which was extrapolated from the radial component $B_{z}$ of the photospheric magnetic field using the Green function method (Sakurai, 1982). Note that pressure and gravity were neglected here because the low $\beta$ approximation was well satisfied in the solar corona (Gary, 2001). Therefore, we solved the MHD equations simplified in the zero $\beta$ approximation. The basic equations are as follows:

where $\rho$, ${\bf B}$, ${\bf v}$, ${\bf J}$, and $\phi$ are plasma density, magnetic flux density, velocity, electric current density, and a convenient potential to remove errors derived from ${\bf\nabla}\cdot{\bf B}$ (Dedner et al., 2002), respectively. In these equations, the length, magnetic field, density, velocity, time, and electric current density are normalized by $L^{*}=244.8Mm$, $B^{*}=2,500G$, $\rho^{*}$ = $|B^{*}|$, $V_{\rm A}^{*}\equiv B^{*}/(\mu_{0}\rho^{*})^{1/2}$, where $\mu_{0}$ is the magnetic permeability, $\tau_{\rm A}^{*}\equiv L^{*}/V_{\rm A}^{*}$, and $J^{*}=B^{*}/\mu_{0}L^{*}$, respectively.

The pseudo density is assumed to be proportional to $|{\bf B}|$ to ease the relaxation by equalizing the Alfv$\acute{e}$n speed in space. $\nu$ is the viscosity fixed at $1.0\times 10^{-3}$, and the coefficients $c_{h}^{2}$, $c_{p}^{2}$ in Equation (5) are also fixed with the constant values, 0.04 and 0.1, respectively. The initial density condition is given by $\rho=|{\bf B}|$, and the velocity is set to zero in the entire space. The resistivity is given by $\eta=\eta_{0}+\eta_{1}|{\bf J}\times{\bf B}||{\bf v}|/|{\bf B}|$ where $\eta_{0}=5.0\times 10^{-5}$ and $\eta_{1}=1.0\times 10^{-3}$ in non-dimensional units. The second term is introduced to accelerate the relaxation to the force free field, particularly in weak field regions. At the boundaries, all of velocities are fixed to zero, and the magnetic fields are fixed to the potential fields except at the bottom boundary. More details are presented in Inoue et al. (2018).

The X2.2 and X9.3 flares occurred in an interval of approximately 3h as shown in GOES⁵⁵5Geostationary Operational Environmental Satellite: We employed GOES-13 because GOES-15, which was the latest GOES series as of September 2017, had brief outage of soft X-ray due to lunar transit near the onset time of the X2.2 flare. soft X-ray light curves of Figure 1. The onset time was 08:57 UT for the X2.2 flare and 11:53 UT for the X9.3 flare, on September 6, 2017. Another C2.7 flare occurred before the two X-class flares, and its onset time was 07:29 UT, on September 6, 2017. Figure 2(a-c) shows the temporal evolution of the radial magnetic field $B_{z}$ from SDO/HMI SHARP. Panels (a), (b), and (c) are snapshots at 04:58:42 UT (approximately 4h before the X2.2 flare), 08:46:42 UT (approximately 10min before the X2.2 flare) and 11:22:42 UT (approximately 30min before the X9.3 flare). We labeled the major polarities as N1 and N2 for the negative region, and P1 and P2 for the positive region, as shown in panel (a). The most remarkable magnetic field change was the intrusion of a part of the N1, which is indicated by the red circle, toward the northwest direction as indicated by the red arrow in panel (a). The region was shaped like a peninsula and intruded into the neighboring positive polarity region P1. Henceforth, this region is called ”intruding negative peninsula.” It was originally shaped as shown in panel (a), and it seemed to start the intrusion between 06:50 and 07:50 UT, when the HMI observation was not available, or the data were not of good quality because of the lunar transit. Then, the intrusion became rapid from around 08:00 UT, i.e., from approximately 1h before the X2.2 flare onset. The intrusion continued throughout the two X-flares (panel (b)). Then, the intruding negative peninsula finally merged with the N2, dividing the P1, as shown in panel (c).

Precursor brightening⁶⁶6We defined precursor brightening as a small brightening intermittently observed at the same site on a magnetic polarity inversion lines where initial flare ribbons appeared on both sides. A physical interpretation of precursor brightening in terms of flare trigger process is discussed in Bamba, & Kusano (2018). and initial flare ribbons of the two X-flares are observed near the intruding negative peninsula. Panels (d-i) are snapshots of the last precursor brightening (panels (d) and (g)), initial flare ribbons (panels (e) and (h)), and enhanced flare ribbons (panels (f) and (i)) for the X2.2 and X9.3 flares, observed by SDO/AIA 1600Å. Several noticeable features were obtained by comparing the X2.2 and X9.3 flares. First, the last precursor brightening was observed at the tip of the intruding negative peninsula before the X2.2 flare, while the location of the last precursor for the X9.3 flare was slightly south from the tip of the intruding negative peninsula, as indicated by the red arrows in panels (d) and (g). In addition, the shape of the last precursor brightening was dot-like for the X2.2 flare, while it was elongated along the magnetic polarity inversion line between P1 and N1 for the X9.3 flare. Second, the initial flare ribbons of the X2.2 flare appeared near the site where the last precursor brightening was observed. The ribbons over the P1 were labeled as R1 and R2, and those over the N1 and N2 were labeled as R3, R4, and R5, as indicated by the white arrows in panel (e). The ribbons R1-R3 were locating around the the tip of the intruding negative peninsula, although the R4 and R5 were at a slight distance from it. The initial flare ribbons of the X9.3 flare also appeared on both sides of the location of the its precursor brightening, although the ribbons were saturated immediately, as seen in panel (h). Third, the enhanced flare ribbons were extended further south for the X9.3 flare, compared with those of the X2.2 flare. Both the X-flares showed north-south (along the magnetic polarity inversion line between P1 and N1) and east-west (along the magnetic polarity inversion line between P1-N2) ribbons, as shown in panels (f) and (i). In the X2.2 flare, the east-west ribbons R4 and R5 appeared almost simultaneously with the north-south ribbons R1-R3; R5 was fainter than the others. Even after the ribbons were enhanced, their shape and location were not very different in panels (e) and (f). In contrast, the X9.3 flare ribbons were extended more toward west and south directions in panel (i) than those in panel (h). These temporal variations in precursor brightening and flare ribbons can be clearly seen in the supplemental animation of the SDO/AIA 1600 Å.

The intruding negative peninsula is likely to related to the onset of the two X-class flares, as described in Section 3.1. Therefore, we further investigated the magnetic field structure and corresponding plasma dynamics near the intruding negative peninsula. Figure 3 (a) shows the three components of magnetic fields in the AR observed by Hinode/SOT-SP, approximately 3h before the X2.2 flare. Hinode/SOT-SP successfully scanned the central part of the AR including the intruding negative peninsula, although it missed the southern half of P2. The magnetic polarity inversion line between P1 and N1, including the intruding negative peninsula, had a strong magnetic gradient in the radial component ${B_{z}}$ and strong magnetic shear in the horizontal components $(B_{x},B_{y})$, as shown in panel (a). The magnetic shear distribution is more clearly shown in panel (b), where a strong magnetic shear ($<-70^{\circ}$ of shear angles) is distributed along the magnetic polarity inversion line between P1 and N1. In particular, the tip of the intruding negative peninsula was close to $-90^{\circ}$ magnetic shear as the vectors of the horizontal fields were almost parallel to the magnetic polarity inversion line in panel (a).

Interestingly, a cusp-shaped brightening was observed at the tip of the intruding negative peninsula approximately 1.5h before the X2.2 flare onset. It was observed as a transient enhancement of soft X-ray as indicated by the red arrow in Figure 1. Figure 4 shows quicklook maps of intensity, Doppler velocity, and line width from Hinode/EIS spectroscopic observations of Fe XIV (264.7Å) line, which is sensitive to emissions from plasma at a temperature of $log(T)\sim 6.3$ with overlying magnetic polarity inversion lines of the radial magnetic field $B_{z}$ from SDO/HMI SHARP. Hinode/EIS field-of-view successfully covered both the north-south and east-west magnetic polarity inversion lines, and the cusp-shaped brightening can be seen in the intensity map of panel (b). The cusp-shaped brightening bridged over the north-south magnetic polarity inversion lines between P1 and N1 at the tip of the intruding negative peninsula. Hinode/EIS scanned the AR from west to east (from right to left in the maps) in the field-of-view for 3min, and the time in the Figure 4 indicates the scan start time. Therefore, the cusp-shaped brightening was observed approximately at 07:19 UT. The brightening was simultaneously observed by SDO/AIA in 171Å, which is sensitive to emissions from plasma at a temperature of $log(T)\sim 5.8$ (Lemen et al., 2012), with considerably higher 12min temporal resolution, as shown in panels (e-i)⁷⁷7Note that data of the AIA 171 Å on panels (e-i) are not good quality because of lunar transit. We could still clearly identify the cusp-shaped brightening and its evolution as shown in panels (e-i).. The brightening started in a small region indicated by the white arrow in panel (e), and propagated to both sides; then, the cusp-shaped brightening appeared (see panels (e-i)). This suggests that there was an energy input at the top of the cusp, and plasma fell down along the magnetic field lines.

Moreover, transient but significant downflow was observed to the west of the intruding negative peninsula, as shown in panel (c). Because Hinode/EIS performed observations with a $2^{\prime\prime}$ slit, the red-shifted Doppler signal was only seen in one slit position. It can be identified that the red-shifted signal corresponds to the west (right in the image, roots in P1) foot point of the cusp-shaped brightening. The Doppler velocity was more than $80km/s$ in the Fe XIV line. Significant line broadening was also observed for the cusp-shape in panel (d). One of the reasons of line broadening is the presence of bi-directional plasma jets (e.g., Innes et al., 1997) or turbulent motion (e.g., Imada et al., 2008) caused by magnetic reconnection. There has been considerable discussion on line broadening associated with magnetic reconnection. The shape of the observed spectral line in our study is similar to the spectral line associated with magnetic reconnection discussed in a previous study (see Figure1 in Innes et al. (1997)). Furthermore, coronal dimming and strong blue-shifted signals were observed in the northeast of the field-of-view in Figure 4(b-d). At 06:21 UT, this coronal dimming had already started and was further enhanced after the X2.2 and X9.3 flares. It is inferred that the coronal dimming related to the expanding coronal loops and corresponding upflows were observed a few days or a day before the flare onset, as suggested by Imada et al. (2014).

We analyzed the spectral line profiles at the site of the cusp-shaped brightening via double Gaussian fitting. Figure 5 shows the spectral line profiles and spectral images of Fe XIV (264.7 Å and 274.2 Å), Fe XVI/Fe XXIII (263.3 Å), and He II (256.3 Å) lines at the single site of the red-shifted signal, shown in Figure 4(c). The vertical axis of the spectral images (a-2, b-2, c-2, and d-2) corresponding to the Y-axis of Figure 4(c), and the white horizontal line-cut denotes the Y-coordinate of the strongest red-shifted signal. The intensity profiles along the white line-cut are plotted in the left for each spectral image (a-1, b-1, c-1, and d-1). The horizontal axis for both intensity profiles and spectral images are converted to denote velocity unit by single (in panels (c) and (d)) or double (in panels (a) and (b)) Gaussian fitting. Evidently, both Fe XIV lines have significant red-shifted components, as shown in panels (a-1) and (b-1). The Doppler velocity of the second components (red-colored profiles) relative to the first components (blue-colored profiles) was approximately $103\pm 8km/s$ for 264.7 Å and $115\pm 16km/s$ for 274.2 Å, respectively. Note that the errors are defined as the summation of errors due to photon noise for each component of the double Double Gaussian fitting. Fe XVI/Fe XXIII (263.3 Å) and He II (256.3 Å) lines in panels (c-1) and (d-1) also show red-shifted profiles although it was not well fitted via double Gaussian fitting. These profiles Fe XVI/Fe XXIII and He II lines were also seen to have red-shifted components with a velocity of approximately $100km/s$.

The global coronal structure of the AR 12673 was observed using Hinode/XRT and SDO/AIA. Figure 6 shows the temporal variation of high-temperature coronal loops observed using Hinode/XRT, in which the magnetic polarity inversion lines of the radial magnetic fields $B_{z}$ from SDO/HMI SHARP or Hinode/SOT-SP are overplotted. The red global coronal loops connect the major bipoles P1-N1, P1-N2, and P2-N1 in panel (a). In particular, P1-N1 loops were highly sheared and almost parallel to the magnetic polarity inversion lines. The small loop that is indicated by the white arrow in panel (a) existed in the north of the intruding negative peninsula at least from 00:00 UT on September 6, 2017. A cusp-shaped brightening then appeared just before the C2.7 flare as seen in panel (b). This cusp-shaped brightening (c.f. Tsuneta et al., 1992) corresponded to the one that was observed by Hinode/EIS and SDO/AIA 171 Å in Figures 4 and 5, and the transient enhancement of soft X-ray as denoted by the red arrow in Figure 1. The cusp-shaped brightening remained until the X2.2 flare onset, following which, a dark tube-like feature was observed just before the X9.3 flare onset, as denoted by the sky-ble arrows in panels (c) and (d). This tube-like dark feature was elongated north-south and was along the magnetic polarity inversion lines between P1 and N1.

Figure 7 shows the coronal loops of lower temperature than those shown in Figure 6, observed by SDO/AIA 171 Å. Panels (a) and (b) are snapshots of HMI radial magnetic field and AIA 171 Å image, respectively, taken approximately 45min before the X2.2 flare onset. There was a small loop that is bridging over the magnetic polarity inversion line between P1 and N1 at the tip of the intruding negative peninsula, as indicated by the red arrow in panel (b). This small loop corresponded to the cusp-shaped brightening seen in Figures 4(e-i) and 6(b). Moreover, another bright loop L1 is denoted by the white arrow in panel (b). The loop L1 remained until the X2.2 flare onset, as seen in panel (c). The initial flare ribbons, shown in Figure 2(e), are also seen in panel (c), although the loop L1 only can be seen in the SDO/AIA 171 Å image. The loop L1 seemed to correspond the coronal loop that was rooted in P1 and N1. The loop L2, which has the same connectivity as L1, appeared after the X2.2 flare, as denoted by the white arrow in panel (d). The intensity of L2 gradually decayed, meanwhile a long north-south loop that is connecting P2-N2, denoted by a white arrow in panel (e), gradually appeared. The long north-south loop expanded as is outlined by white dashed line in panel (f) and erupted simultaneously with intense emission of the X9.3 flare onset.

Interestingly, the dark tube-like structure that is shown in Figure 6 (c) and (d) is also observed in the AIA 171 Å image, as denoted by the sky-ble arrow in Figure 7 (e). This tube-like structure is along the magnetic polarity inversion lines between P1 and N1 and is also observed under the bright loop L2 in panel (d). It was not observed in AIA 171 Å image before the X2.2 flare; however, it started to be seen as gradually rise from around 10:45 UT, when the brightening of the X2.2 flare ribbons was decaying as shown in the supplemental animation. The dark tube-like structure then became invisible because of intense emission from the X9.3 flare ribbons in panel (f). It seemed to be erupted by the X9.3 flare because there was no clear structure such as dark tube after the X9.3 flare emission started decaying.

The X-class flares occurred while the intruding negative peninsula was moving toward northwest as we described in Section 3.1. Our results suggest that the intrusion possibly causes variations in magnetic topologies owing to magnetic reconnection along the magnetic polarity inversion line between P1 and N1. Figure 8(a) shows the magnetic field lines that produce the X2.2 flare as blue and magenta tubes on the radial magnetic field image. The positive/negative foot point of magenta (blue) tubes is labeled as F1/F2 (F3/F4), as shown in panel (b), in which the back ground is the image of the initial flare ribbons in AIA 1600 Å. The negative foot points of magenta (F2) and blue (F4) tubes are rooted in intruding the negative peninsula (i.e., N1) and N2 respectively, while the positive foot points of both tubes (F1 and F3) are rooted in P1. Note that a part of the positive foot point of the blue tubes is rooted slightly closer to the intruding negative peninsula although still in P1; thus, we labeled the foot point as F’3.

The negative foot point F2 of the magenta tubes is pressed against the blue tubes as the intruding negative peninsula moves toward northwest. This motion possibly causes “push-mode” magnetic reconnection (Yamada, 1999) that forms a magnetic flux rope erupting with the X2.2 flare. Figures 9(a) and (b) illustrate the before and after versions of push-mode magnetic reconnection that varies magnetic topologies from F1-F2 /F3-F4 (the magenta/blue tubes) to F2-F3/F1-F4 (yellow tubes). It is inferred that F1-F4 develops as a flux rope and the bright loop L1 in Figures 7 (b) and (c) corresponds to overlying magnetic fields on the magnetic flux rope. The overlying loop L1 becomes bright as the flux rope grows and rises toward its eruption, i.e., the X2.2 flare, owing to the intrusion of the intruding negative peninsula. The initial flare ribbons then appear at each foot point as shown in Figure 2 (e). The foot points F1 and F3 in Figure 8 (b) correspond to the initial flare ribbons R1 and R2 on the positive polarity P1. The other ribbons R3, R4, and R5 on the negative polarities N1 and N2 appear at the foot points of F2 and F4. Here, the foot point F4 is rooted by both the flare ribbons R4 and R5.

Moreover, F’3-F4, which is a part of the pre-existing blue tubes and is denoted by the blue dashed line in Figure 9(a), is also reconnected with the magenta tubes F1-F2. F’3-F4 reconnects with F1-F2 earlier than F3-F4 because it is originally located closer to F1-F2 than F3-F4. This magnetic reconnection with F’3-F4 and F1-F2 forms a small loop F2-F’3, which is denoted by the red dashed line in Figure 9(b), and corresponds to the small cusp-shaped brightening that is observed in Hinode/EIS, XRT, and SDO/AIA 171 Å (cf. Figures 4, 6, and 7) before the X2.2 flare.

Figure 8(c) displays the magnetic field lines before the X9.3 flare. The intrusion of the negative peninsula continues even after the X2.2 flare, and it further causes magnetic reconnection under the yellow tubes. A part of yellow tubes then becomes unstable and is lifted. Thus, the initial flare ribbons of the X9.3 flare appear under the yellow tubes as shown in Figure 8 (d). The yellow tubes further reconnect with the green tubes that are rooted in P2. This magnetic reconnection causes variations in the magnetic topologies from F1-F4/F5-F6 (the yellow/green loops) to F4-F6/F1-F5 (thick and thin red loops), as illustrated in Figures 9(c) and (d). Furthermore, the initial flare ribbons propagate to the P2 region, where F6 is rooted in, and are elongated toward the south more than that of the X2.2 flare, as shown in Figure 2(i).

The magnetic flux rope F1-F4, which is indicated by the thick yellow line in Figure 9(c), might correspond to the dark tube-like structure that was observed in the Hinode/XRT and SDO/AIA 171 Å images between the X2.2 and X9.3 flares. The tube-like structure started to observed in SDO/AIA 171 Å from 10:45 UT (approximately 1.5h before the X9.3 flare onset), while it was not seen in SDO/AIA 171 Å and Hinode/XRT before and just after the X2.2 flare. Furthermore, we observed precursor brightening that is elongated along the magnetic polarity inversion line between P1 and N1 before the X9.3 flare onset, as denoted by the red arrow in Figures 2(g) and 7(e). Those observational results suggest that successive magnetic reconnection under the flux rope F1-F4 occur and reinforce the flux rope (cf. Fan, 2010).

Therefore, according to the results obtained from observational data analysis, the intrusion of the negative peninsula that presses the magenta magnetic field lines to the blue magnetic field lines promote push-mode magnetic reconnection that causes the X2.2 flare. Moreover, the successive intrusion of the negative peninsula further plays an important role to cause successive magnetic reconnection and to hasten the eruption of the yellow magnetic flux rope by an MHD instability. Zou et al. (2020) also stated the importance of the first X2.2 flare reconnection to facilitate the rise of the flux rope and its eruption, i.e., the X9.3 flare, by torus instability. The magnetic polarity inversion line between P1 and N1 has strong magnetic gradient, resulting in strong electric current. Therefore, there is a possibility that magnetic reconnection between sheared magnetic fields over the P1-N1 magnetic polarity inversion lines occurs and forms a magnetic flux rope by consuming a considerable amount of time, even without the intrusion of the negative peninsula. Nevertheless, our results suggest that the intrusion of the negative peninsula encourages the formation and destabilization of magnetic flux ropes both in the X2.2 and X9.3 flares.

In this study, we found local and transient downflow related to the cusp-shaped brightening as we described in Section 3.2. Generally, Lorentz force driven flows such as reconnection out flow do not exhibit clear temperature dependence while pressure gradient driven flows such as chromospheric evaporation exhibit the strong temperature dependence (e.g., Fisher et al., 1985; Watanabe et al., 2010; Imada et al., 2015). We observed the downflow of $\sim 100km/s$ at both high temperature Fe lines ($log(T)\sim 6.3-6.4$) and low temperature He II line ($log(T)\sim 4.9$). The temperature is believed to rises proportionately to height in the solar atmosphere in the region where is higher than the chromosphere (e.g., Vernazza et al., 1981). Therefore, each ion’s sensitive temperature is suggested to correspond to the formation height of each emission line. Further, the downflow illustrate in Figure 4(c) is driven by Lorentz force and that there is an energy input, such as magnetic reconnection, that occurred in a higher atmosphere than formation height of the Hinode/EIS Fe lines shown in Figure 5. The velocity of the downflow was approximately $100km/s$, which is significantly lower than the Alfv$\acute{e}$n velocity (order of $1,000km/s$) in the corona, in all the Fe and He lines shown in Figure 5. The small cusp-shaped high-temperature loop that was observed using Hinode/XRT in Figure 6(b) and magnetic topology in Figure 8 also supports this interpretation.

As we discussed in Section 4.1, the push-mode magnetic reconnection between the magenta and blue sheared magnetic fields due to the intrusion of the negative peninsula can form a small magnetic loop connecting P1 and N1 at the tip of the intruding negative peninsula, as well as the cusp-shaped brightening. Moreover, downflow was observed in the west leg of the cusp rather than the top of the cusp, as seen in Figure 4(b, c). Therefore, it is suggested that the downflow indicates a secondary flow in which coronal plasma decelerate and fall down along the coronal loop (e.g., Imada et al., 2013; Polito et al., 2018), rather than the reconnection out-flow itself (e.g., Yokoyama, & Shibata, 1998).

Bamba et al. (2017) found similar local and transient upflow using lower temperature lines ($log(T)\sim 4.0-4.8$), which were observed by the IRIS⁸⁸8Interface Region Imaging Spectrograph satellite (De Pontieu et al., 2014), than the Hinode/EIS observed lines in this study. In their case, temperature independent strong upflow with a velocity of up to $100km/s$ was observed, corresponding to a brightening in the chromosphere before an X-class flare. They interpreted that the local and transient upflow suggest the presence of an energy input from magnetic reconnection in a lower temperature region lower than the middle chromosphere. Unfortunately, data from IRIS, which can observe emission lines from a lower temperature atmosphere such as that used in Bamba et al. (2017), was not available for this AR 12673. Even so, it is suggested that our downflow is driven by Lorentz force because both the high-temperature Fe lines and low-temperature He II line showed red-shifted signals with the velocities of approximately $100km/s$.

Moreover, in Bamba et al. (2017), the local and transient upflow, which was observed approximately 30min before an X-class flare onset by IRIS, is reported as a proxy of precursory phenomenon that directly represents the trigger of an X-class flare. Their upflow was very transient but the corresponding brightening in the chromosphere continued until 10min before an X-class flare onset. In our case, it seems that the location of the cusp-shaped brightening and downflow corresponds to the location of the push-mode magnetic reconnection, which plays a crucial role in triggering the X2.2 flare. However, it is unreasonable to assume that those features represent the direct trigger of the X2.2 flare because those were observed more than 1.5h before the flare onset. Therefore, it is suggested that the cusp-shaped brightening and downflow are proxies of the push-mode magnetic reconnection that gradually occurs at the tip of the intruding negative peninsula due to its intrusion from before the X2.2 flare. As we described in Section 4.1, the magnetic connectivity of the cusp-shaped brightening seems to be formed by push-mode magnetic reconnection between the magenta and blue magnetic arcades in Figure 8 (a) and (b). This topological change possibly forms a magnetic flux rope corresponding to the F1-F4 loop and is erupted with the X2.2 flare.