Imaging Evidence for Solar Wind Outflows Originating from a Coronal Mass Ejection Footpoint

We focus on the eruption of a long filament F1. The event is introduced in Figure 1, which shows the field of view (FOV) of AIA in the 304 Å filter channel data before (panel (a)), during (b), and after (c) the eruption. An extensive quiet-Sun region is located in the east, with three filaments F1, F2, and F3 (panel (a)). In the neighborhood of the eruption lie a coronal hole and the NOAA 12333 active region (panels (a) and (b)).

Filament F1 had been forming for nearly two Carrington rotations since early 2015 March. Its eruption began after the eruption of the westernmost filament F3, which took place at ≈3:00 UT on 2015 April 28. The stability of filament F2 did not seem to be altered between the eruptions of F1 and F3. The sympathetic nature of this eruption, if any, is however out of the scope of this study, and we therefore do not analyze F2 and F3 any further.

Due to its orientation and large size, filament F1 cannot be traced along its whole extent. Using the standard solar coordinates, its southern leg was rooted in an area corresponding to $X\approx -400^{\prime\prime}$ and Y ≈ 200''–400''. Its northern footpoints were not visible because they were either obscured by the filament itself or located behind the solar limb. F1 lay above the PIL, which we roughly indicate using a yellow line in Figure 1(d).

Panels (e)–(g) of Figure 1 show the eruption as viewed by the LASCO C2 coronagraph. In panel (e), the apex of the filament, visible as a bright blob above the occulter, is overlaid by an arcade of long coronal loops. This envelope is being pushed outward by the erupting filament, which is depicted in panel (f). There, its two legs can be distinguished, one rooted toward the north and the other toward the south. The legs are more visible in panel (g), in which most of the erupting filament has already left the FOV. Panels (f)–(g) also reveal the presence of a crossing of the filament's legs, visible between the filament and the occulter, creating a so-called writhed elbow (Zhang et al. 2012; Cheng et al. 2013b). The eruption of F1 was not accompanied by a clearly identifiable flare; the GOES hard X-ray flux was, during the whole studied period, below the C-class threshold and did not show a discernible peak (panel (h); see also Section 6).

The eruption of F1 was a very long–lasting event, since eruption-related phenomena were observed for more than 24 hr (see Section 4.3). Filament F1 started to erupt after ≈12:00 UT, although motions of the threads composing its southern leg could be discerned even at earlier times. To specify the onset time of the eruption, we constructed a time–distance diagram along the direction of the eruption (white arrow in Figure 2(a)). F1 first rose slowly for about 2 hr up until ≈12:00 UT (Figure 2(b)), when it exponentially accelerated. To distinguish between the slow- and fast-rise phases of the eruption, we fitted the motion of the leading edge of the filament using the fitting function of Cheng et al. (2013a):

$$\begin{eqnarray}&&h(t)={c}_{0}{e}^{(t-{t}_{0})/\tau )}+{c}_{1}(t-{t}_{0})+{c}_{2}.\end{eqnarray} \tag{ 1 }$$

There, ${c}_{0},{c}_{1},{c}_{2},{t}_{0}$, and τ are free parameters whose initial estimates were taken from McCauley et al. (2015, Table 3 therein). The parameter c₁, handling the increment of the linear term, was found to be very small, of the order of 10⁻⁴, with an uncertainty of the same order of magnitude. We have therefore discarded the linear term and fitted the filament using the exponential and constant terms only. The resulting fit is shown using a blue dashed line in Figure 2(b). Separate linear fits (solid white lines) were then used to obtain the velocities of rising during the slow- and fast-rise phases of the eruption. These were found to be ≈1 km s⁻¹ and 126 km s⁻¹, respectively.

Zoom In Zoom Out Reset image size

Since we were not capable of calculating the onset time using the full analytic formula as in McCauley et al. (2015), for its estimation we used the absolute differences between the linear fit of the slow-rise phase (extended using the white dotted line in Figure 2(b)) and the exponential fitting function. We found that these functions exceed the threshold corresponding to the spatial resolution of AIA after ≈12:10 UT. Similar onset time can also be obtained via expanding the fitting function into the Taylor polynomials.

Note that this fit of the leading edge of the filament is not representative of the motion of the individual threads composing it. As can be seen in Figure 2(b), at the beginning of the eruption, some threads slightly descended between ≈10:00 and 11:30 UT, most likely due to twisting motions of filament material during the eruption. Apart from this, faint threads can be distinguished both in front of (green arrows) and behind the filament (red arrows). The transition between the slow- and fast-rise phases for these threads occurred at various instants between ≈12:30 and 13:40 UT.

During the eruption, two flare ribbons formed and at least one of them developed a J-shaped (hooked) extension (Figure 1(c)). In order to investigate which magnetic polarities were occupied by the ribbons, we manually traced the ribbons using the 304 Å filter channel data and replotted them over HMI ${B}_{\mathrm{LOS}}$ data saturated to ±30 G (red dashed lines in Figure 1(d)). Since the eastern ribbon was spatially coincident with regions of negative magnetic polarity, we will refer to it as the negative-polarity ribbon. Analogically, we will refer to the conjugate hooked ribbon rooted in positive magnetic polarity as the positive-polarity ribbon and to its hook as "the hook."

The evolution of the flare ribbons and in particular that of the hook are here first traced manually using the 304 Å filter channel data. In this channel, the ribbons are well visible and unobstructed by coronal emission. The positions of the positive-polarity ribbon and the hook at different times are shown in Figure 3(a). The hook started to form roughly after 14:00 UT, during the fast-rise phase of the eruption (Section 3.1). We were first able to distinguish its curved part (elbow) located in the south. Both its ends then elongated toward the northeast (along the PIL), one extending to what became the tip of the hook and the other toward the straight part of the positive-polarity ribbon. At this point, the entire hooked ribbon became visible as a single structure (black line). The hook then started to contract and, together with the rest of the ribbon to which it was attached, to drift toward the north (blue, green, and red lines).

The hook can also be observed in the coronal filter channels. The 171 and 193 Å filter channel data are shown in panels (c)–(h) at a 2 hr cadence. Animated versions of these observations are available online. In the 171 Å filter (panels (b)–(d)), the quiet-Sun network surrounding the hook can be seen. Panels (e)–(h) showing the 193 Å filter reveal the quiet-Sun region with numerous funnel-shaped loop footpoints. The relatively darker region seen in the southern part of panel (e) is the hook in its early stages of formation. Later, after the hook elongated (panel (f)), we see that some of the funnel-shaped loop footpoints were located at the inner edge of the hook—for example, at ≈[50'', 180'']. For simplicity, these will further be referred to as the funnels. Apart from the funnels, low-lying bundles of quiet-Sun loops are seen to surround the hook. Later, as the hook drifted toward the north, some of these structures disappeared (panels (g)–(h)), presumably due to the magnetic reconnection at the ribbon hook. This phenomenon is further analyzed in Section 6.2.

After the eruption, a coronal dimming formed in the area encircled by the hook (Figures 3(f)–(h)). The intensity drop was caused by the disappearance of the individual structures rooted in this region, mainly of the funnels, which progressively narrowed (panels (b)–(d) and (f)–(h)) until finally fading away after 21:00 UT. There were also multiple funnels rooted in the northern part of the dimming region, which faded within 2 hr of their formation (panels (e)–(f)). Besides, a few small-scale TR structures were observed in both the 171 and 193 Å filters. Most of them disappeared before 20:00 UT, as shown in Figure 3.

Funnels located at the inner edge of the hook are projected to lean over the dimming region, which naturally enhances their contrast. We investigate the funnel marked using the cut c1 in Figure 3(f), because it is both the largest and the brightest one and is visible for the longest period. The time–distance diagrams constructed along this funnel are shown in Figure 4. They contain full-cadence data imaged in the 193, 171, and 304 Å filter channels of AIA. To increase the signal-to-noise ratio, they were smoothed with a 3 × 3 pixel boxcar, which roughly corresponds to the spatial resolution of the instrument.

Zoom In Zoom Out Reset image size

The time–distance diagram showing the 193 Å data is first dominated by a bright region at the position $c1\lessapprox 25^{\prime\prime}$. This emission originates in the lower portions of the funnel and fades after 14:20 UT. Roughly at the same time, the funnel seems to slightly brighten along the whole extent of c1. This phase remains present until ≈15:00 UT, when vertical stripes, repeated in a riblike pattern, start to be visible along most of c1. The inclination of these stripes indicates the presence of a motion directed along c1. Given the orientation of c1, this implies that the material is moving from lower to higher altitudes. This phenomenon lasts for more than 5 hr until ≈20:20 UT. During this time, the funnel gradually fades until it disappears (see Figure 3). In the remainder of this section, we will assume this motion to be a signature of outflows, while a discussion on its nature is left for Section 5.

The period shown in Figure 4 can, to a first approximation, be divided into three different episodes, named the "pre-outflow," "onset," and "outflows and dimming" phases. They are named according to the variations of the brightness of the funnel and the presence of outflows and are highlighted using colored lines above Figure 4.

In order to find the instants limiting the commencement and termination of each phase, we produced lightcurves using the 193 Å channel data averaged along c1. The normalized lightcurves, smoothed over 10 minute intervals, are shown in Figure 5. The lightcurves were produced along the upper part of the cut, $c1\gt 60^{\prime\prime}$, and along its lower part, $c1\lt 40^{\prime\prime}$. This choice excludes the bright underlying TR that is projected at positions $c1\approx 40^{\prime\prime} \mbox{--}60^{\prime\prime}$ along the cut. In the same figure we also plot the function fitting the rise of the filament (gray line) derived in Figure 2(a).

Zoom In Zoom Out Reset image size

The dashed line, showing the intensities averaged at low altitudes ($c1\lt 40^{\prime\prime}$), decreases for most of the analyzed period. This decrease is typically slow except for the period between ≈13:30 UT and ≈15:30 UT, when the averaged intensity drops by about 40%. This darkening corresponds to the formation of the hook and the dimming region within. In contrast, the intensity averaged near the tip of c1 ($c1\gt 60^{\prime\prime}$, solid line) first slightly increases and after ≈12:30 UT reaches a plateau. After ≈14:00 UT it rises again, peaks at ≈14:30 UT, and starts to drop. The averaged intensity reaches the initial pre-rise value at ≈15:00 UT. These three instants are highlighted in Figure 5 using blue diamond, red triangle, and green cross symbols. Note that after the peak of the solid curve, both lightcurves decrease in a similar manner. This implies that the funnel and the outflows along it were dimming evenly along their whole extent.

It is difficult to find the exact time of the onset of the outflows using Figure 4 only, as the transition is gradual and the event is very long–lasting. We will therefore refer to the period of increased brightness, between the blue diamond and the green cross, as the transition phase between the "pre-outflow phase" and the phase characterized by the presence of outflows ("outflows and dimming"). The transition phase is co-temporal with the fast-rise phase of the eruption (Section 3.1) and the formation of the dimming region (Section 3.4). For simplicity, we will refer to this period as the outflow "onset" phase.

Apart from their gradual fading, the pattern generated by the outflows in Figure 4 as well as their inclination does not change during the third phase. To measure the velocities of the outflows, we fitted multiple stripes observed approximately between ≈15:00 UT and 19:40 UT with linear fits whose endpoints were selected manually. The resulting velocities, listed in Table 1, range between ≈70 and 140 km s⁻¹, while the mean velocity is 112 km s⁻¹. Their uncertainties were estimated using Equation (1) of Dudík et al. (2017). There, the spatial resolution of AIA was adopted as the uncertainty in the spatial position ${\sigma }_{{\rm{s}}}$. The uncertainty in time (${\sigma }_{{\rm{t}}}$) was, by measuring the FWHM of the fitted stripes, estimated as 54 s.

Table 1.  Outflow Velocities and Velocity Uncertainties Measured in the Dimming Region, Resulting from Different Fits

#	$v\pm {dv}$ (km s−1)	#	$v\pm {dv}$ (km s−1)
0	73 ± 8	5	122 ± 26
1	109 ± 16	6	111 ± 20
2	124 ± 21	7	122 ± 26
3	89 ± 15	8	119 ± 29
4	114 ± 23	9	140 ± 37

Download table as:  ASCIITypeset image

The outflows can also be seen in the time–distance diagram produced using the 171 Å filter channel data. During the pre-eruption phase, the TR structures under c1 are seen to emit at positions $c1\approx 40^{\prime\prime}$, ≈55'', and $\gt 80^{\prime\prime}$. Apart from that, an irregular V-like pattern at $c1\lt 25^{\prime\prime}$ can be seen. It is caused by a swaying motion of the TR structures near the footpoints of the funnels and lasts until ≈19:00 UT. The outflows can easily be distinguished after 14:00 UT. The outflows share the same pattern and inclination of those seen in the 193 Å diagram, but after ≈19:00 UT, their intensity decreases more slowly.

The outflows cannot be found in the time–distance diagram produced using the 304 Å channel data. It only reveals the V-like pattern in the TR, similar to the one observed in the 171 Å data. Besides, vertically stretched features at $c1\approx 40^{\prime\prime} \mbox{--}60^{\prime\prime}$ and $c1\gtrapprox 80^{\prime\prime}$ can be seen. Most of them fade away in the outflow phase.

We now compare the outflows in the dimming region to the outflows along funnels rooted in a proper coronal hole located in the vicinity. In its southern part, three funnels can be distinguished in both the 171 and 193 Å filter channel data (Figure 3). Their characteristics are studied using time–distance diagrams, which we constructed along the funnel highlighted using the cut c2 (Figure 3(f)).

The time–distance diagram constructed using the 193 Å filter channel data is shown in Figure 6. Since the eruption had no apparent effect on the morphology of the funnels in the coronal hole, the period included therein is shortened compared to the one shown in Figure 4. However, to facilitate comparison between the time–distance diagrams constructed in the coronal hole and the dimming region, the 1 hr intervals on the X-axis correspondingly have the same plotted length.

Zoom In Zoom Out Reset image size

At positions $c2\lessapprox 70^{\prime\prime}$, vertical, inclined stripes indicating outflows can be distinguished during the whole period. When compared to the outflows seen in the hook, they seem to be sharper and their contrast is higher, i.e., they can be recognized more easily. The stripes are steeper and narrower, and the corresponding ${\sigma }_{{\rm{t}}}$ is 48 s only. Even though their mean velocity of 135 km s⁻¹ is slightly higher than that in the hook, the individual velocities listed in Table 2 are, within the uncertainty, comparable to those reported on in the hook. We note that the outflows can also be seen in time–distance diagrams produced using the 171 Å filter channel data (not shown). They are however more difficult to distinguish.

Table 2.  Outflow Velocities and Velocity Uncertainties Measured in the Coronal Hole, Resulting from Different Fits

#	$v\pm {dv}$ (km s−1)	#	$v\pm {dv}$ (km s−1)
0	168 ± 64	5	127 ± 37
1	113 ± 25	6	139 ± 53
2	146 ± 40	7	150 ± 49
3	109 ± 45	8	109 ± 36
4	154 ± 76	9	131 ± 40

Download table as:  ASCIITypeset image

Note that apart from the outflows, a few bright blobs are seen at $c2=40^{\prime\prime} \mbox{--}50^{\prime\prime}$. Again, their emission corresponds to TR structures emitting from underneath c2.

During and after the eruption, the positive-polarity ribbon hook with the dimming region within slowly drifted toward the coronal hole located to the north. This convergence is shown in Figure 7, where contours corresponding to 35 DN s⁻¹ pix⁻¹ were plotted to highlight the boundaries of the hook and the coronal hole. As seen in panels (b)–(c), this convergence was accompanied by the fading of low-lying bundles of quiet-Sun loops rooted in the relatively narrow region between the dimming region and the coronal hole. We note that even though their morphology is similar to that of the funnels located in the dimming region, no outflows were observed to occur along them. For simplicity, these will further be referred to as the canopies.

Zoom In Zoom Out Reset image size

The following day at ≈06:00 UT (panel (d)) the canopies separating the hook and the coronal hole disappeared, evidencing a merging of these two structures. Positions corresponding to the hook selected by hand at this time are also indicated in Figure 3(a) using a red line. There, the opening of the hook's northern part toward the coronal hole is evident. After ≈16:00 UT on April 29, traces of the hook can no longer be distinguished.

Note that this observation of the merging of a dimming region with an ordinary coronal hole is not unique. For example, this phenomenon can be seen to follow the multiple filament eruptions studied by Gutiérrez et al. (2018; see Figures 2(d)–(g) therein).

As mentioned in Section 1, imaging observations evidencing the presence of outward-oriented plasma motions from dimming regions have not been reported so far. These motions are however often observed in active regions, where they are referred to as propagating intensity, or coronal, disturbances (see, e.g., the review of De Moortel 2009). Even though these structures are morphologically distinct from dimmings, their edges are often blueshifted with outflow velocities corresponding to those found in dimming regions (Section 1). For example, the pattern visible in Figure 4 is reminiscent of that reported by Sakao et al. (2007; see Figure 2(b) therein). Their observations were interpreted as outflows contributing to the slow solar wind. The same event was revisited, e.g., by Nishizuka & Hara (2011), who discovered that the properties of the outflows change with height, suggestive of the presence of upward-propagating waves along the outflows. The debate about the origin of these disturbances continued in numerous publications; for a thorough review we refer the reader, e.g., to De Moortel & Nakariakov (2012). An attempt to resolve the dispute about the origin of the propagating disturbances has been carried out, e.g., by De Moortel et al. (2015). Even though they introduced a method based on observations of Doppler perturbations as a probe to distinguish between wave- and flow-like interpretation, they concluded that distinguishing between the two might not be possible at all.

Concerning the eruption analyzed in this manuscript, since the material was propagating along the funnel-shaped loops (i.e., along the field lines), the waves acting there would most likely be Alfvén waves. If we, for simplicity, take ${B}_{\mathrm{LOS}}$ to range within 1–30 G, which can be taken as the upper limit of ${B}_{\mathrm{LOS}}$ (Figure 1(d)), and assume typical quiet-Sun densities of $(2\mbox{--}3)\times {10}^{8}$ cm⁻³ (e.g., Del Zanna & Mason 2018; Lörinčík et al. 2020), we obtain Alfvén speeds of the order of 10²–10³ km s⁻¹. Their lower limit might thus correspond to the velocities of the upward-propagating brightenings in the dimming region (Table 1). Some of their properties however turn us to favor the outflow scenario.

First, the variations of B(z) and ${N}_{{\rm{e}}}(z)$ with height would lead to variations of ${v}_{{\rm{A}}}$ along c1, which we did not observe. Second, the characteristics of MHD waves should change in time. Their amplitudes undergo damping (De Moortel & Hood 2003), caused mainly by thermal conduction and viscosity (Mandal et al. 2016; Ruan et al. 2016). Changes in their periodicity have also been evidenced (Bryans et al. 2016). The properties of the outflows we report on however do not seem to change during the whole observation, as the inclination of the stripes visible in Figure 4 does not change for at least 5 hr. Third, even if we neglect the damping and changes of periodicity, ${v}_{{\rm{A}}}$ would increase in time due to the drop in density during the evolution of the dimming (see, e.g., Cheng et al. 2012; Vanninathan et al. 2018; Hou et al. 2020). The reported drops of density reaching up to 70% would cause ${v}_{{\rm{A}}}$ to progressively increase by a factor of up to ≈1.7, which is again contradicted by our observations.

In the framework of the standard solar flare model and its 3D extensions, outflows in the dimming region hosting the legs of the erupting flux rope (in our case filament F1) should occur due to the expansion of the erupting structure, which leads to the elongation of the legs of the CME. The relatively long period during which the outflows were observed during this event however suggests that they are not related to a stretching-induced dilution of plasma only. Combining the estimated height of the erupting filament (≈150 Mm) with the measured speed of the CME (126 km s⁻¹; Section 3) results in a dilution timescale of ≈20 minutes only. This estimate corresponds, within the order of magnitude, to the duration of the brief intensity decrease visible after the onset phase in Figure 5. Still, it is a factor of ≳15 shorter than the duration of the outflow phase in the dimming region. This means that outflows in dimming regions persisting for such large timescales are likely to be powered by a different mechanism.

The similarity between the characteristics of the flow-like motions observed in the dimming region and the coronal hole suggests that we were observing solar wind–like outflows. As the dimming region started to form, the funnel-shaped loop footpoints turned to true coronal funnels, known as one of the sources of the solar wind (e.g., Marsch 1999; Tu et al. 2005; Marsch et al. 2008; He et al. 2010). Possibly, the opening of the magnetic field during the CME led to a geometry favoring a free flow of plasma toward the interplanetary space in the same manner as that in the coronal holes. As we have shown, the hook merged with the coronal hole, at which point the two became indistinguishable (Figure 7). Note however that this merging was not co-temporal with the observations of the outflows in the dimming region, as it continued for many hours after the outflows disappeared (Section 4.3).

The 3D ar–rf reconnection describes how arcades ("a") overlying the flux rope reconnect with its leg ("r") into new lines composing the flux rope ("r") and flare loops ("f"). It involves the hook; during its expansion, its outer edge sweeps arcades, which become a part of the erupting flux rope. They swap their connectivities with field lines composing the flux rope, swept by the inner portion of the hook. Consequently, these then turn to flare loops. The sequential change of connectivity between the arcades and the field lines composing the flux rope results in an apparent drifting of flux rope footpoints along the hook.

Simultaneous observations of all constituents of the ar–rf reconnection are rare, as not all reconnecting structures are visible, particularly in the case of dimmings. One-to-one correspondence between the observed and the predicted manifestations has been reported in one event only (Dudík et al. 2019). Its indications however have been reported in a few studies. The evolution of the flare ribbons analyzed by Zemanová et al. (2019) was found to be consistent with ar–rf reconnection predictions. Field lines composing the erupting hot channel were however not distinguished. Similar results were achieved by Chen et al. (2019), who reported on the drift of the flux rope footpoints based on the motion of flare kernels. Lörinčík et al. (2019b) observed a conversion of filament threads into flare loops accompanied by a disappearance of arcades rooted near the hook. Drifting of the filament threads toward the tip of the hook was however not observed. The authors speculated that some of the post-reconnective filament field lines might have reconnected into the dimming region located in the hook and therefore were not observed.

As mentioned in Section 3.3, the leg of the filament was, before the eruption, rooted in the quiet-Sun region with multiple funnel-shaped loop footpoints. These can be divided into two groups: those hosting the outflows (funnels, Section 4.1) and those that were not (canopies, Section 4.3). Some of the canopies located to the northwest of the dimming region were swept by the outer edge of the forming hook. This process is detailed in the first column of Figure 8. Panel (a) shows a view of the region where the hook formed. Using a blue circle we marked a canopy that we selected for further analysis. During the elongation of the hook (panel (b)), the tip of the hook converged toward this canopy. Twenty minutes later (panel (c)), after the hook elongated toward the northeast, the portion of the canopy located near the edge of the hook disappeared. At 16:40 UT (panel (d)), when the hook further elongated in the same direction, the entire circle was a part of the dimming region formed in its interior. The canopy thus became a part of the erupting filament, resulting in a spatial expansion of the dimming region.

The second row of Figure 8 shows the evolution of the ribbon hook as well as the threads composing the erupting filament in the 304 Å filter channel data (panels (e)–(h)). The threads are difficult to track, mainly because they are dark and projected over the gradually forming dimming region. Therefore, to highlight their motion, we produced log-running ratio (LRR) images using the same 304 Å data. The time delay between the two frames producing one LRR image was chosen as 30 minutes, which was found to best highlight the moving thin threads over the evolving dimming region (panels (i)–(l)). In addition, to further enhance the visibility of the threads, the LRR images were smoothed with the 3 × 3 boxcar, while both the original and the LRR images were averaged in three consecutive exposures. We also encourage the reader to review the animated version of these panels, where the motion of the filament threads is easy to discern.

In the LRR images, the filament threads are at each time highlighted using pink arrows. Their footpoints are in both the 304 Å and the LRR, marked using white–pink circles. At 13:20 UT (panels (e) and (i)) the threads composing the filament can be traced all the way to their footpoints located approximately at [0'', 120''] (circle 1). Later at 13:40 UT, the footpoints have drifted slightly toward the northwest (panels (f) and (j), circle 2). Because of the bright material falling along the threads, they are again observed to drift toward the west 1 hr later in panels (g) and (k) (circle 3). Afterwards (panels (h) and (l)), the footpoints of the threads are no longer distinguished. The threads however shift toward the north, which is consistent with the understanding that the footpoints, though not visible, drifted further along the hook.

This drift of the filament threads and their footpoints is a clear manifestation of the drift of the CME legs in the same way as in Dudík et al. (2019). Together with the evolution of the hook (Figure 3(a)), it confirms the presence of the recently identified 3D reconnection ar–rf geometry (Aulanier & Dudík 2019), which leads to the drifting of the flux rope footpoints in hooked flare ribbons due to their reconnections with the surrounding solar corona.

The distribution of hot plasma was investigated using the 94 Å filter channel data. There was only a small amount of hot emission associated with this event; this is supported by the flat curve of the GOES X-ray flux, which did not rise during the eruption (Figure 1(h)). To enhance the emission of flare loops and increase their signal-to-noise ratio, we averaged the data in 20 consecutive exposures and smoothed them with the 3 × 3 boxcar. The resulting images are shown in Figures 8(m) and (n). There, flare loops joining the negative-polarity ribbon to the positive-polarity ribbon and its hook are found, although they are still faint with average intensities usually below 1 DN s⁻¹ pix⁻¹. These low intensities explain why there are no flare loops clearly identifiable in individual 94 Å exposures nor a discernible GOES peak.

Flare loops can first be distinguished at around 18:00 UT, approximately 3 hr after the formation of the hook, and are rooted in its elbow (as in Lörinčík et al. 2019b). In panels (m) and (n) we indicate the initial positions of the filament's footpoints using circle 1. As time went by and the hook drifted toward the north, these locations started to correspond to the footpoints of flare loops (highlighted using yellow arrows). This evidences a conversion of field lines composing the erupting filament into flare loops, which is in agreement with the predictions imposed for ar–rf reconnection.

Although we could not find all four structures that mutually change the connectivities during 3D ar–rf reconnection, the present observations provide significant evidence that ar–rf reconnection took place.

First, the canopies surrounding the hook were swept by its outer edge, i.e., $a\to r$ (arcade to flux rope) conversion took place. Second, we observed the drift of the filament footpoints during the eruption. Third, the previous position of the filament footpoint turned to flare loops ($r\to f$). All this is piecewise consistent with the predictions of the ar–rf reconnection geometry, and the latter cannot be explained by the traditional reconnection present in the 2D CSHKP model, where only arcades reconnect to become a flux rope and a flare loop. We note that so far, the majority of eruption events with hooked ribbons reported in literature (Aulanier & Dudík 2019; Lörinčík et al. 2019b; Zemanová et al. 2019) have allowed us to discern only partial constituents of the ar–rf reconnection geometry, while providing valuable evidence that field lines indeed reconnect in this newly identified 3D reconnection geometry. The present event is no exception.

Our observations indicate that the canopies rooted near the tip of the hook became a part of the erupting filament via the ar–rf reconnection. This means that closed loops reconnect to become a part of the dimming region. Thus, their new magnetic field geometry is one favorable for the development of outflows as shown in Section 4. By introducing more and more field lines into the dimming region, the ar–rf reconnection led to its spatial expansion.

In contrast to that, the funnel-shaped loop footpoints, used for the analysis of the outflows (cut c1, Section 4.1), did not enter the dimming region via the ar–rf reconnection. As the dimming region formed during the eruption, these structures were already rooted at its inner edge (Figures 3(f)–(g)) and were therefore not reconnecting. Instead, their transformation to true coronal funnels, with outflows, was driven by the opening of the field lines in the dimming region (Section 5.2).

Along this funnel, the outflows were observed for many hours irrespective of the ar–rf reconnection occurring between ≈13:20 and 14:50 UT near the tip of the hook. The funnel itself however did not reconnect up until ≈21:00 UT, when it was swept by the drifting hook (Figure 3(h)), which is a signature of magnetic reconnection (see Aulanier & Dudík 2019). Interestingly, the fact that the funnel was swept by the hook and not the straight part of the ribbon shows that the funnel was a part of the erupting flux rope.