PLASMOID EJECTIONS AND LOOP CONTRACTIONS IN AN ERUPTIVE M7.7 SOLAR FLARE: EVIDENCE OF PARTICLE ACCELERATION AND HEATING IN MAGNETIC RECONNECTION OUTFLOWS

We reconstructed RHESSI X-ray images in energy bands from 3 to 50 keV using the CLEAN algorithm and detectors 3–9. Depending on the count rate, we chose variable integration time ranging from 20 s during the impulsive phase to 4 minutes during the decay phase.

Figure 3 and its accompanying movie show examples of RHESSI contours overlaid on AIA images. There is a persistent loop-top source at 6–10 keV (green) near the apex of the cusp-shaped EUV loops. Accompanying the evolution of the EUV loops mentioned above, this loop-top source undergoes a gradual ascent followed by a descent and then another ascent. This can be best seen in the last panel showing the temporal migration of the emission centroid obtained from a contour at 70% of the maximum of each image. Early descents of loop-top X-ray sources have been recognized (e.g., Sui et al. 2004), but this is the first time that an evident preceding ascent is observed.

Zoom In Zoom Out Reset image size

Early in the event there is an additional, weaker coronal X-ray source (panel (b)) located in the lower portion of the overlying flux rope. It later falls below detection with the eruption of the flux rope. During the impulsive phase (panel (e)) at higher energies (25–50 keV), there are double footpoint sources and a Masuda-type (e.g., Masuda et al. 1994; Nitta et al. 2010) coronal source located 20'' = 15 Mm above the SXR (6–10 keV) loop top. This separation is twice that of the Masuda case.

The earlier C4.5 flare (panel (a)), though more compact, exhibits surprisingly similar emission, including the additional X-ray source within the flux rope and the Masuda-type hard X-ray (HXR) source. This suggests that the end state of the first, confined flare of a failed eruption serves as the initial state of the second, eruptive flare, making them homologous flares. Note that the weaker, southern footpoint emission is (partially) occulted by the limb, especially for the first flare.

To track various moving features, we placed eight cuts of 10'' wide centered on the limb, as shown in Figure 2(f). Cut 0 is positioned along the presumable current sheet between the initial flare cusp and erupting flux rope, and is used as the fiducial line for obtaining projected heights up to ∼08:00 UT. Subsequent cuts are evenly spaced by 3° to capture the cusp at different stages as it gradually turns toward the south. For each cut, we averaged pixels of an AIA image sequence within it in the perpendicular direction to obtain a space-time plot.

Figure 4(c) shows, for example, a base-ratio (i.e., normalized by the initial intensity profile) space-time plot from Cut 0 at 131 Å. It shows the eruption of the flux rope and the three-stage development (upward, downward, and upward again) of the underlying flare cusp. The peak emission at each time, marked by the small purple symbols, evidently exhibits this motion. Space-time plots of other AIA channels are shown in Figure 13 and described in Appendix A.

Zoom In Zoom Out Reset image size

Figure 4(e) shows the height–time history of the loop-top emission centroids in four RHESSI X-ray bands from 3 to 25 keV and in all AIA EUV channels. (From now on, we refer to both X-ray centroids and EUV peaks as centroids.) We find that all loop-top centroids follow the same three-stage motions. The AIA 131 and 94 Å channels best exhibit this trend continuously, while in cooler channels, especially 211 and 171 Å, the initial ascent and subsequent descent become obscure. Note that the 3–6 keV centroid (red diamonds) cannot be identified since the impulsive phase because the RHESSI thin attenuator moves in and raises the energy threshold to 6 keV.

We also notice a clear energy dispersion that the loop-top emissions at higher temperatures or photon energies are located at greater heights, with all X-ray centroids situated above EUV centroids. For example, at its greatest height prior to the impulsive phase, the 16–25 keV centroid is 35'' = 26 Mm above its 335 Å counterpart. This energy dispersion is consistent with previous observations (e.g., Veronig et al. 2006) and in line with the expected picture that higher loops are newly energized and thus are hotter and/or host nonthermal electrons of harder spectra, while lower loops are previously energized and have undergone cooling. We suggest that the so-called above-the-loop-top Masuda-type sources, including the 25–50 keV coronal source in Figure 3(d), could be extreme cases of this general trend.

In Table 2, we list the initial heights of X-ray and EUV loop-top centroids around 04:35 UT, together with the maximum and minimum heights during their descents and the percentage height reductions. In general, the above noted energy dispersion persists and the 10%–20% descents are in agreement with earlier reports (Veronig et al. 2006). An exception is the 193 and 304 Å channels because of their response to both hot and cool emissions (see Appendix A). There is a weak trend that higher energy X-ray descents start (at the maximum heights) later, but still within 5 minutes during 05:02–05:07 UT, and X-ray descents end (at the minimum heights) earlier than EUV descents by up to 10 minutes.

We measured the average velocities of the loop-top centroids by fitting a piecewise linear function to the height–time data during their initial ascent, subsequent descent, and second ascent stages. As listed on the right-hand side of Table 2, the result shows a general trend of higher velocities at higher energies. The X-ray descents of −11 to −15 km s⁻¹ are about twice faster than the EUV descents except for 193 Å because of its dual temperature response. There is also a gradual decrease in velocity with time during the second ascent stage through the decay phase. These trends generally agree with previous observations (Liu et al. 2004; Veronig et al. 2006).

To follow the temporal variations of the loop-top velocities, we took time derivatives of spline fits to the centroid heights of selected channels. The resulting velocities versus time are shown in Figure 4(c) for the 6–10 and 10–16 keV and 131 and 94 Å channels. The outstanding feature is that the maximum descent velocities (also tabulated in Table 2) in the range of −7 to −23 km s⁻¹ are all attained during 05:15–05:20 UT, near the onset of the impulsive phase at 05:16 UT marked by a vertical dashed line. This time is delayed from the initial descent by about 10 minutes. This indicates that the loop-top descent velocity, rather than its height, is more intimately correlated with the energy release rate.

We also show in Figure 4(e) the centroid heights of the upper coronal X-ray source located in the lower portion of the flux rope. A parabolic fit indicates a final velocity of 19 km s⁻¹ at 04:50 UT, comparable to or slower than those in other events (Liu et al. 2008, 2009c; Sui & Holman 2003). It is also 50% slower than the middle portion of the then accelerating flux rope, as indicated by the parabolic fit in Figure 13(e), which reaches 153 ± 3 km s⁻¹ later at 05:07 UT near the edge of AIA's field of view (FOV).

The conjugate footpoints at 25–50 keV during the impulsive phase move away from each other at an average velocity of 13 km s⁻¹ (see Figure 4(e)), almost identical to the velocity of the simultaneous loop-top ascent (see Table 2). This is consistent with previous observations (Liu et al. 2004) and the classical picture of eruptive flares during the arcade growth phase.

Figure 6 shows kinematic measurements of the slow loop shrinkages. These loops traverse a hot (up to T ∼ 30 MK; see Figure 11) loop-top region, which is bright only in the hottest channels of 193, 131, and 94 Å, and descend toward the apex of the flare arcade seen in cooler channels, where they fade below detection (see Figure 13). They generally start at a low velocity with a median of −17 km s⁻¹ (+/− for upward/downward) and decelerate to typically −5 km s⁻¹ within 0.5–2.5 hr. Their median initial deceleration is 0.014 km s⁻².

Zoom In Zoom Out Reset image size

Such shrinkages are persistent throughout the flare, not only during the decay phase when the overall loop-top emission develops upward as previously observed (Forbes & Acton 1996), but also during the pre-impulsive and impulsive phases when the loop top undergoes its initial ascent and subsequent descent. Fifty shrinkages are identified during 04:00–16:00 UT with an average occurrence rate of once per 7.7 minutes up to 10:00 UT but more frequent around the impulsive phase (Figure 6(b)).

Regardless of the direction of motion of the loop-top emission, the initial height of loop shrinkage generally increases with time. The initial velocity v₀ exhibits more variability, as shown in Figure 6(c), and appears to be positively correlated with the flare energy release rate indicated by the time derivative of the GOES flux (black dotted line). In particular, near the onset of the impulsive phase at 05:16 UT (vertical dashed line), v₀ increases rapidly and reaches a maximum of −58 km s⁻¹. There is a similar temporal correlation with the initial deceleration a₀ (panel (d)) that also generally correlates with v₀.

Figure 7 shows, in the same form as Figure 6, kinematic measurements of fast downward loop contractions as well as upward plasmoid ejections that will be examined in Section 4.3. We color-code fits to tracks of different categories: upward ejections in red and downward contractions in cooler colors. For the latter, we use green for early contractions that occur simultaneously with upward ejections and originate from lower heights h < 100'', and blue for later contractions that occur without observed upward counterparts (possibly out of AIA's FOV, after the flux rope has erupted) and originate from greater heights h > 100''. We identified 29 green tracks and 174 blue tracks during 04:00–16:00 UT.

Zoom In Zoom Out Reset image size

As shown in Figure 7(c), the fast downward contractions, especially those later blue-colored ones, start well above the hot loop-top region and travel into it with deceleration for some distance before fading below detection. Their final heights are generally above the original heights of slow loop shrinkages (see Figure 6(b)).

In original AIA images, the fast contracting features are usually bright, cusp-shaped loops (see Figure 2, bottom), but occasionally dark, tadpole-like voids (McKenzie & Hudson 1999; Innes et al. 2003; Cassak et al. 2013). In running ratio space-time plots (Figure 7, top), the former each produce a bright track followed by a recovering dark track, while the latter produce tracks of reversed order. We treat them all as contracting loops and do not distinguish their differences that are beyond our scope.

As shown in Figure 7(c), the early, green-colored contractions have short ranges of 10''–20'', because of the compactness of the flare at this time, while the later, blue-colored contractions travel great distances up to 200'' = 150 Mm. However, their statistical medians, modes, and means of the initial velocities are quite similar within 200–300 km s⁻¹ (Figures 7(d) and (e); Table 3), suggestive of a common origin of these contractions possibly as reconnection outflows. The initial decelerations of the early contractions are, on the other hand, about twice higher than those of the later ones, with a median of 2.022 versus 0.848 km s⁻² (Figures 7(f) and (g)).

The later, high-altitude contractions have a mode velocity of −158 km s⁻¹ that is similar to the median velocity of −150 km s⁻¹ of supra-arcade downflows (Savage & McKenzie 2011; Warren et al. 2011). However, their median and mean velocities, −242 and −282 km s⁻¹, are nearly twice greater. This is due to the high-velocity tail with 36 out of 174 (or 21%) contractions being faster than −400 km s⁻¹ (hatched area in Figure 7(e)), reaching a maximum of −918 km s⁻¹. We ascribe this difference to AIA's improved capabilities that allow us to detect such contractions in their early stages when their velocities are high and emissions are weak.

The initial velocities and decelerations of the earlier and later fast contractions, when taken together, generally increase with time up to ∼06:10 UT and then decrease. This maximum time is delayed by almost 1 hr from that of the slow loop shrinkages and the onset of the HXR burst represented by the GOES flux derivative shown in Figure 6(c). The temporal variations here seem to be more closely correlated with the GOES flux itself. The late-phase decreases of the initial velocities and decelerations are physically reasonable as energy release subsides, but there is an observational bias for those tracks starting at the edge of AIA's FOV which may have decelerated before they are first detected.

Figure 8 shows the distributions of fractional height reductions of the downward loop shrinkages and contractions, defined as (h₀ − h_f)/h₀, where h₀ and h_f are the initial and final projected heights. The slow shrinkages (light blue) and later, high-altitude fast contractions (blue) have similar distributions with medians of 0.3 and 0.26, respectively, which are comparable to those observed in SXRs (Forbes & Acton 1996; Reeves et al. 2008) and predicted theoretically (Lin 2004). The early, low-altitude fast contractions (green) have a median of twice smaller, but comparable to the 10%–20% descent of the overall loop-top emission (Table 2). Meanwhile, the later fast contractions have a sizable fraction of 16% (27 out of 174) with height reductions more than 0.4 (up to 0.7), which is larger than predicted. Note that these observed values are lower limits, because loops, especially those detected at the edge of AIA's FOV, could continue contraction before and after their observed intervals when their emission remains below detection.

Zoom In Zoom Out Reset image size

As shown in Figure 7, upward plasmoid ejections appear as steep, elongated tracks (red fits). Compared with the downward loop contractions (green fits) at the same time, they travel greater distances (20''–80'' versus 10''–20''), with larger initial velocities (median: 322 versus −208 km s⁻¹) but smaller decelerations (median: −0.018 versus 2.022 km s⁻²). (The opposite signs of the acceleration and velocity here indicate deceleration.) Some upward ejections even experience acceleration. Such differences, especially the ratio of their median velocities of 208/322 = 2/3, are consistent with previous observations and simulations of oppositely directed reconnection outflows (Takasao et al. 2012; Bárta et al. 2008; Shen et al. 2011; Murphy et al. 2012; Karpen et al. 2012). These differences are likely due to different environments: the upflow runs into a region of (partially) open field lines with low plasma density, while the downflow impinges on high density, closed flare loops that can cause strong deceleration.

The starting heights of these upward ejections and their downward counterparts (Figure 7(c)) both increase with time. This suggests an upward development of the reconnection site situated between the opposite outflows (e.g., Shen et al. 2011). We estimated the height of the reconnection site from a spline fit, shown as the green dashed line in Figures 7(a)–(c), to the initial heights of the bi-directional fast outflows. Its initial velocity of a few  km s⁻¹ is similar to that of the initial loop-top ascent, suggesting the upward development of reconnection being its underlying mechanism. This estimated reconnection site leaves the AIA FOV at about 06:40 UT.

During 04:15–05:25 UT, we identified 25 upward ejections (red) and 29 downward, early fast contractions (green), at average occurrence rates of once every 2.4 and 2.2 minutes, respectively. These rates are comparable to that of the later fast contractions (blue, up to 10:00 UT) of once every 2.0 minutes and to those in MHD simulations (Bárta et al. 2008).⁴ The persistence throughout the entire flare of such contractions also agrees with the statistical result from Yohkoh (Khan et al. 2007).

The initial velocity of upward ejections, as shown in Figure 7(d), increases with time and seems to be correlated with the height of the overlying flux rope as it evolves into eruption. In particular, there is a rapid velocity increase at the onset of the impulsive phase, reaching 1050 km s⁻¹. This aspect of this event is independently studied by Liu (2013).

To infer the overall properties of nonthermal particles and thermal plasma, we analyzed spatially integrated RHESSI spectra of individual detectors following the procedures detailed in Liu et al. (2008) and Milligan & Dennis (2009).

Figure 9 shows example spectra in three phases of the flare. (1) During the pre-impulsive phase (04:59:00–05:01:32 UT), we fitted the spectrum with two isothermal functions combined: a warm component (red dotted) of temperature T = 13 MK and emission measure EM = 0.37 × 10⁴⁸ cm⁻³ plus a hot component (green dotted) of 30 MK (also called super-hot; Caspi & Lin 2010) and a 40 times smaller EM. If we were to replace the hot component with a power-law ∝E^−γ, we would have a steep slope of γ = 7.6. (2) During the impulsive phase (05:25–05:26 UT), we used an isothermal (T = 22 MK, EM = 2.8 × 10⁴⁸ cm⁻³) plus power-law (blue dotted, γ = 3.8) model. (3) During the decay phase, the data can be fitted with one isothermal function of a slightly reduced temperature of 18 MK but twice higher EM of 5.5 × 10⁴⁸ cm⁻³. Note that at E > 25 keV, the power law in the impulsive phase is more than an order of magnitude higher than the thermal component. This indicates that the 25–50 keV Masuda-type loop-top source shown in Figure 3 is primarily nonthermal and likely a particle acceleration region itself (e.g., Liu et al. 2008; Krucker et al. 2010).

Zoom In Zoom Out Reset image size

We applied such spectral fits throughout the flare and the temporal variations of the fitting parameters are shown in Figures 1(d) and (e). In general, the warm thermal component is persistent in time, and both its T and EM (orange plus signs) increase through the HXR burst followed by a more gradual decrease. A similar trend is present for the hot thermal component (red crosses) during the pre-impulsive phase. A thermal fit to GOES data gives a somewhat lower T but higher EM (black dotted lines), because of its well-known preferential response to relatively cooler plasma. The power-law component (blue) displays a common hardening trend before the RHESSI night data gap.

In Section 3.2 we have examined the energy-dependent height distribution of the loop-top X-ray source (lower coronal source). Here we extend this analysis to the upper coronal source during the pre-impulsive phase and compare the two sources together.

As shown in Figures 10(a) and (b), the higher-energy emission (12–15 keV, blue contours) of the lower source is located at higher altitudes than lower-energy emission (3–8 keV, green contours), while the upper source has an opposite trend. That is, the two sources are closer to each other at higher energies, as can be better seen in panel (d) from the linear fits to their centroid heights as a function of logarithmic energy. This implies higher temperatures of the thermal plasma and/or harder spectra of the nonthermal particles within the inner regions. Similar RHESSI observations have been reported (Sui & Holman 2003; Liu et al. 2008, 2009c) and interpreted as evidence of magnetic reconnection and associated energy release being located between the double coronal sources.

Zoom In Zoom Out Reset image size

At the highest energies (≳10 keV), this trend seems to be reversed, especially for the upper source. In previous observations (e.g., Figure 4 in Liu et al. 2008), more obvious reversals occurred at slightly higher energies of 15–20 keV and were interpreted as a transition from the low-energy thermal regime to the high-energy nonthermal regime, in which greater stopping distances of higher energy electrons can cause higher energy bremsstrahlung X-rays being located farther away. We further suggest that this reversal may be an indication of the two X-ray sources being spatially separated, instead of merging together at the highest energies; so are the locations of primary heating and particle acceleration from magnetic reconnection. This is supported by the separate temperature peaks at the two sources as shown in Figures 11(a) and (h) and discussed in Section 5.3.

Zoom In Zoom Out Reset image size

To infer the spatial distributions of temperature T and emission measure (EM), we employed a forward-fitting algorithm using AIA filter ratios (Aschwanden et al. 2013; for a different approach, see Battaglia & Kontar 2012). This algorithm assumes a Gaussian differential EM that has a peak EM at temperature T. Such a model provides the simplest description of the EM-weighted temperature distribution along the line of sight.

Figure 11 (left) shows examples of T and EM maps. Early in the event at 04:49 UT, the highest temperature of log T = 6.95 or T = 8.9 MK is located in the rising flux rope and the underlying flare cusp, but EM is highly concentrated in the latter. The temperature then increases with time especially in the flare cusp, reaching log T = 7.45 or T = 28.2 MK at 05:09 UT, close to the 35 MK of the hot RHESSI component at this time (see Figure 1(d), red crosses). In the decay phase, the highest temperatures are located in the outer layer of the flare arcade and the lower portion of the contracting cusps (ray-like features) above it. The EM in the latter is nearly four orders of magnitude lower than in the central arcade, similar to those of hot fans of rays observed by Yohkoh (Švestka et al. 1998). The considerable increase of EM and thus density in the flare loops from panel (d) to (e), prior to the impulsive phase, suggests chromospheric evaporation driven by thermal conduction (Liu et al. 2009b; Battaglia et al. 2009) or by Alfvén waves (Haerendel 2009), rather than electron beam heating that may be important later in the presence of HXR footpoints.

To better follow the history of T and EM, we obtained their space-time plots, as shown in Figures 11(g) and (k), from corresponding maps using Cut 0. We identified the maximum temperature and EM at each time and show their projected heights as orange and black symbols in panel (p). In general, both peaks follow the same upward–downward–upward motions as the loop-top emission centroids. The temperature peaks are near the RHESSI 16–25 keV centroids (blue), while the EM peaks are close to or slightly lower than the AIA 131 Å centroids (purple) at lower heights.

The peak offset of the temperature and EM can be better seen in their height distributions at selected times shown in Figures 11(h) and (l). It explains the observed energy dispersion of the X-ray loop-top centroids shown in Figure 4(e) because of the exponential shape of thermal bremsstrahlung spectra (Equation (4)). As shown by the green and red lines in the middle of Figure 9, a higher temperature but lower EM produces a harder (shallower) spectrum of lower normalization that dominates at high energies, while a lower temperature but higher EM produces a softer (steeper) spectrum of higher normalization that dominates at low energies. The temperature peak being located above the EM peak thus shifts the higher energy X-rays toward greater heights. Otherwise, if the temperature and EM peaks are cospatial, they would dominate X-rays at all energies and there would be no separation of centroids with energy. We note in Figure 11(h) that the EM decreases more gradually near the flux rope. This may lead to the less pronounced energy dispersion of the upper coronal source there than the lower (loop-top) source (see Figure 10).

As a proof of concept, we modeled the observed X-rays of energy E with thermal bremsstrahlung radiation (Tandberg-Hanssen & Emslie 1988, p. 114):

$$\begin{equation} I_{\rm SXR} \propto ({\rm EM}) {\exp (-E/kT) \over E \sqrt{T} } g(E/kT), \end{equation} \tag{ 4 }$$

where EM is the emission measure and g(E/kT) = (kT/E)^2/5 is the Gaunt factor. The resulting X-ray profiles at two selected times are shown in Figures 11(i) and (m) for the corresponding T and EM profiles. As expected, the higher energy 20 keV emission (blue) is dominated by the temperature peak at a higher altitude, while the 3 keV emission (red) is dominated by the EM peak at a lower altitude. Their emission peaks, marked by open circles, differ in height by Δh = 7'' and 46'' for the two times. Their peak heights are repeated in panel (p) and are close to those of observed loop-top centroids.

As shown in Figures 11(g) and (p), the high-temperature region and particularly the temperature peak are close to the X-ray loop-top centroids but always below the reconnection site (green dashed line). This indicates that primary plasma heating takes place in reconnection outflows. To illustrate this, we can follow a contracting loop and the temperature variation it senses as it travels away from the reconnection site. A selected track from Figure 7(c) is shown here as the blue dotted arrow. The temperature history on its path, as shown in Figures 11(h) and (j), indicates rapid heating at an average rate of 3.2 MK minute⁻¹ from 3.5 to 28 MK over 8 minutes and Δh = 90 Mm. This example gives us a sense of the heating rate averaged along the line of sight, not necessarily of a specific loop.

Likewise, the track of a slow loop shrinkage at lower altitudes reveals cooling at an average rate of −0.32 MK minute⁻¹ from 25 to 7.9 MK within 54 minutes. The rate is −1 MK minute⁻¹ earlier during the impulsive phase in the 28–9 MK range and −0.1 MK minute⁻¹ later during 07:00–08:30 UT in the 14–7 MK range. These cooling rates are somewhat lower than previously found at even lower temperatures (Vršnak et al. 2006).

We have presented detailed EUV and X-ray observations of the 2012 July 19 M7.7 flare, focusing on the signatures of magnetic reconnection and associated energy release. We summarize our findings and discuss their implications as follows.

• 1.  
  The V-shaped EUV emission on the trailing edge of the flux rope CME and the underlying, inverted V-shaped flare loops, both associated with distinct X-ray emission, suggest two oppositely oriented Y-type null points with a vertical current sheet formed in between (Figures 2 and 5).
• 2.  
  Originating from the inferred magnetic reconnection site within the current sheet are bi-directional outflows in the forms of plasmoids and cusp-shaped loops. The upward ejections have a median initial velocity of 320 km s⁻¹ and a maximum of 1050 km s⁻¹, while the concurrent downward cusp contractions are ∼2/3 slower with a median of −210 km s⁻¹. Even faster contractions up to −920 km s⁻¹ occur after the flux rope eruption for another 10 hr, and a sizable fraction of 21% of them have speeds ⩾400 km s⁻¹, twice faster than previously reported (Savage & McKenzie 2011). Such high velocities are comparable to expected coronal Alfvén speeds of ∼1000 km s⁻¹ and sound speeds of 520–910 km s⁻¹ for a 10–30 MK flaring plasma. At lower altitudes, flare loops persistently shrink at a typical initial velocity of −17 km s⁻¹ and gradually decelerate to about −5 km s⁻¹ within 0.5–2.5 hr. They are all evidence of reconnection outflows during different stages (Figures 6 and 7).
• 3.  
  The double coronal X-ray sources are spatially separated and located in the regions of bi-directional plasma outflows. The highest temperature is found near the loop-top X-ray source well below the reconnection site. This suggests that primary plasma heating and particle acceleration take place in the reconnection outflow regions (Holman 2012), rather than at the reconnection site itself. Models with this ingredient were proposed long ago (e.g., Forbes & Priest 1983), but solid observational evidence as presented here has been lacking (Figures 10 and 11).
• 4.  
  An energy dispersion is present in the loop-top position with higher-energy X-ray and hotter EUV emission located at greater heights over a large range of Δh ⩾ 26 Mm. The 25–50 keV nonthermal emission lies 15 Mm (twice that of the Masuda flare) above the 6–10 keV thermal emission. This agrees with the expected trend of softer electron spectra in the nonthermal regime and lower temperatures in the thermal regime being associated with earlier energized loops, which are located further below the primary locus of energy release. The upper coronal X-ray source has an opposite trend because it is located in the oppositely directed reconnection outflow (Figures 3, 4, and 10).
• 5.  
  Prior to the recently recognized descent followed by a continuous ascent, the overall loop-top emission experiences an initial ascent for nearly an hour. This is the first time that such motions, including the descent, are observed simultaneously from EUV to HXRs covering a wide range of temperatures of 1–30 MK. The transition from ascent to descent coincides with the rapid acceleration of the flux rope CME. The loop-top descends at ∼10 km s⁻¹, about 50% slower in EUV than in X-ray (Figure 4).
• 6.  
  The flare impulsive phase starts when rapid velocity increases occur for the overall loop-top descent, the individual loop shrinkages, and the upward plasmoid ejections. This is delayed by 10 minutes from the initial loop-top descent, implying that the energy release rate is more intimately correlated with these velocities than the loop-top position (Figure 4).

We propose the following physical picture to tie together the observations. The major points, including the location of particle acceleration and the three-stage (up–down–up) motion of the loop-top X-ray source, are sketched in Figure 12.

Zoom In Zoom Out Reset image size

The earlier confined C4.5 flare leads to the formation of a flux rope and cusp-shaped loops underneath it. After 6 hr of slow evolution, the flux rope becomes unstable and rises (Patsourakos et al. 2013). A vertical current sheet forms between two Y-type null points at the lower tip of the flux rope and the upper tip of the underlying cusp. Magnetic reconnection ensues within the current sheet, leading to the eruptive M7.7 flare.

Magnetic reconnection produces bi-directional outflows in the forms of the observed plasmoids and contracting loops. Plasmoids are flux tubes formed by the tearing mode (Furth et al. 1963) with a guiding field along the current sheet. They are magnetic islands in two dimensions or when the current sheet is seen edge-on. These outflows are driven by the magnetic tension force of the highly bent, newly reconnected field lines, as in those pointed cusps. The outflows generally decelerate, as observed here, when they run into the ambient corona and when the contracting loops relax to less bent shapes with reduced magnetic tension. This implies that the low-altitude slow shrinkages could be the late stages of decelerated high-altitude fast contractions.

Several mechanisms can operate in the reconnection outflows and contribute to particle acceleration and plasma heating. Turbulence or plasma waves, for example, can be generated by the interaction of the high-speed flows with the ambient corona. Upon cascading to smaller scales (comparable to the gyro-radii of background particles) at some distance from the reconnection site, turbulence can accelerate the particles and heat the plasma (Hamilton & Petrosian 1992; Miller et al. 1996; Chandran 2003; Petrosian & Liu 2004; Petrosian et al. 2006; Jiang et al. 2009; Fleishman & Toptygin 2013). Additional contribution to particle acceleration and/or heating can come from fast-mode shocks in the outflow regions (Forbes & Priest 1983; Tsuneta & Naito 1998; Guo & Giacalone 2012; Nishizuka & Shibata 2013), gas dynamic shocks within contracting flux tubes (Longcope et al. 2009), and the first-order Fermi and betatron mechanisms within the collapsing traps formed by contracting loops (Somov & Kosugi 1997; Karlický & Kosugi 2004; Karlický 2006; Grady et al. 2012).

As the flux rope evolves into its fast rise and eruption stage, the current sheet could become sufficiently long causing significant tearing instability (e.g., Bárta et al. 2008), and/or become thinner than the ion skin depth leading to a transition from Sweet–Parker reconnection to collisionless Hall reconnection (Cassak et al. 2006). Both can result in enhanced rates of reconnection and energy release. The outflow velocity will also increase, producing stronger heating and particle acceleration by one or a combination of the above three mechanisms (turbulence, shock, and betatron) whose efficiency is expected to be positively correlated with the outflow velocity. For example, faster outflows can generate stronger turbulence that can lead to stronger particle acceleration (see Petrosian 2012, for a comparison between acceleration by turbulence and shocks). This explains the observed temporal correlation of the increase in outflow velocity and the onset of the impulse phase and HXR burst.

The loop-top source seen in X-ray and EUV represents the collective emission of the ensemble of flare loops, each of which undergoes post-reconnection contraction or shrinkage. The loop-top emission centroid is the average position of this ensemble, which in the thermal regime is determined by the convolution of the spatial distributions of temperature and EM. Such distributions are influenced by several competing processes, including energization of new loops produced by the upward developing reconnection, and downward contractions and cooling of previously reconnected loops. We suggest that the interplay of these processes results in the up–down–up migration of the loop-top source. Around the early impulsive phase, the observed higher velocities of loop contractions tend to shift the overall loop-top source downward, as depicted in Figures 12(c) and (d). Within the contracting loops, conductive cooling can be considerably reduced by stronger turbulence produced by faster reconnection outflows (Chandran & Cowley 1998; Jiang et al. 2006). Slower cooling helps the X-ray/EUV emission of these hot loops last longer during their contractions and further contribute to the loop-top descent. Before and after the early impulsive phase, the situation could be different. Loop contractions are slower and associated faster cooling can make these loops cool below the instrumental temperature passband more rapidly. The upward development of reconnection could thus dominate, leading to the upward loop-top migration. This interpretation is supported by the observed temporal coincidence of the highest velocities of individual loop shrinkages and of the loop-top descent at the impulsive phase onset. Like cooling, chromospheric evaporation can also play a role. After the initial chromospheric evaporation associated with energization of newly reconnected loops, if it continues operating during the subsequent loop contractions, the densities and thus EMs of these loops would keep rising and contribute to the downward loop-top drift. Otherwise, if it rapidly subsides, it would assist the upward loop-top drift. Investigating such a complex interplay would require detailed numerical modeling, which is beyond our scope here.

The fast downward contractions of EUV loops are best seen during the decay phase, likely because of their greater distances traveled making them easier to be detected. Likewise, different environments may explain the factor of two difference in the median deceleration between the early and later fast contractions (Figures 7(f) and (g)). The earlier, compact flare loops of stronger magnetic field can produce stronger resistance (e.g., by the Lorentz force from a reverse current; Bárta et al. 2008) to quickly brake the contracting loops impinging on them from above, while the later, large flare arcade of weaker field at greater heights and greater travel distances may allow more gradual deceleration.

The X-ray spectra are essentially thermal during the decay phase. This suggests that the late phase contractions are associated with plasma heating rather than particle acceleration. This is expected because the magnetic field strength decreases with height, and so does the available magnetic energy to be released by reconnection. Such prolonged heating may contribute to the well-known slower than expected cooling of flare plasma (see Figure 1(d); McTiernan et al. 1993; Jiang et al. 2006).

The fast high-altitude contractions and slow low-altitude shrinkages appear as two statistically distinct populations. The median final velocity of the former is more than four times the median initial velocity of the latter (−73 versus −17 km s⁻¹; Figures 6 and 7). We have interpreted the latter as the late stages of the former that have decelerated, but we rarely see a continuous track decelerating from >400 km s⁻¹ to a few  km s⁻¹. Alternatively, this distinction may be due to inhomogeneity of reconnection (see, e.g., Asai et al. 2004) in the current sheet above the flare arcade that is primarily oriented along the line of sight. The persistent slow shrinkages could result from reconnection at a moderate rate throughout the current sheet, which produces flare loops forming the arcade and double ribbons at their footpoints. The episodic fast contractions could be signatures of plasmoids within the current sheet, associated with enhanced reconnection and energy release rates. These fast contracting loops are dispersed along the line of sight amidst slowly shrinking loops, producing the observed effects (e.g., Figure 7(b)).

There are other physically different (though not necessarily unrelated) phenomena that share similar observational signatures and should not be confused with the contractions of newly reconnected loops studied here. For example, pre-existing coronal loops can contract at typical speeds <100 km s⁻¹ during eruptions (Liu & Wang 2010; Liu et al. 2012a; Sun et al. 2012), interpreted as implosion (Hudson 2000; Janse & Low 2007) due to the rapid release of magnetic energy and thus reduction of magnetic pressure within the eruption volume. When a CME eruption initiates at an elevated height, it can produce a downward push to displace ambient coronal loops (e.g., Liu et al. 2012c, see their Figure 12(h), at −60 km s⁻¹). Even slower shrinkage at −3 km s⁻¹ can occur in AR loops due to gradual cooling (Wang et al. 1997). Sometimes, conjugate X-ray footpoints approach each other while the loop-top source descends (Ji et al. 2006; Liu et al. 2009a; Yang et al. 2009), likely because of reconnection progressing toward less sheared loops at lower altitudes. We have not found these signatures in the flare under study here, except for a few overlying existing loops that contract at about −150 km s⁻¹ during 05:40–05:50 UT near the end of the impulsive phase. In addition, some high-lying loops continue to expand and erupt at ∼100 km s⁻¹ up to 08:00 UT, 3 hr after the original flux rope eruption. Examination of such features and quantitative comparison of these observations with particle acceleration models will be subjects of future investigations.