The Astrophysical Journal, 923:142 (16pp), 2021 December 20 https://doi.org/10.

3847/1538-4357/ac2ba1
© 2021. The American Astronomical Society. All rights reserved.

Partial Eruption, Confinement, and Twist Buildup and Release of a Double-decker

Filament
1
Jialin Chen1,2, Yingna Su1,2 , Rui Liu3,4 , Bernhard Kliem5 , Qingmin Zhang1 , Haisheng Ji1,2 , and Tie Liu1,2
Key Laboratory of Dark Matter and Space Astronomy, Purple Mountain Observatory, CAS, Nanjing, Jiangsu 210023, People’s Republic of China; [email protected]
2
Department of Astronomy and Space Science, University of Science and Technology of China, Hefei, Anhui 230026, People’s Republic of China
3
CAS Key Laboratory of Geospace Environment, Department of Geophysics and Planetary Sciences, University of Science and Technology of China, Hefei, Anhui
230026, People’s Republic of China
4
Collaborative Innovation Center for Astronautical Science and Technology, Hefei, Anhui 230026, People’s Republic of China
5
Institute of Physics and Astronomy, University of Potsdam, D-14476 Potsdam, Germany
Received 2020 October 28; revised 2021 September 26; accepted 2021 September 28; published 2021 December 16

Abstract
We investigate the failed partial eruption of a filament system in NOAA AR 12104 on 2014 July 5, using
multiwavelength EUV, magnetogram, and Hα observations, as well as magnetic field modeling. The filament
system consists of two almost co-spatial segments with different end points, both resembling a C shape. Following
an ejection and a precursor flare related to flux cancellation, only the upper segment rises and then displays a
prominent twisted structure, while rolling over toward its footpoints. The lower segment remains undisturbed,
indicating that the system possesses a double-decker structure. The erupted segment ends up with a reverse-C
shape, with material draining toward its footpoints, while losing its twist. Using the flux rope insertion method, we
construct a model of the source region that qualitatively reproduces key elements of the observed evolution. At the
eruption onset, the model consists of a flux rope atop a flux bundle with negligible twist, which is consistent with
the observational interpretation that the filament possesses a double-decker structure. The flux rope reaches the
critical height of the torus instability during its initial relaxation, while the lower flux bundle remains in stable
equilibrium. The eruption terminates when the flux rope reaches a dome-shaped quasi-separatrix layer that is
reminiscent of a magnetic fan surface, although no magnetic null is found. The flux rope is destroyed by
reconnection with the confining overlying flux above the dome, transferring its twist in the process.
Unified Astronomy Thesaurus concepts: Solar filament eruptions (1981); Solar flares (1496); Solar magnetic fields
(1503); Solar evolution (1492); Solar extreme ultraviolet emission (1493)
Supporting material: animations

1. Introduction et al. 2006; Liu et al. 2018b). Only part of the filament structure
has erupted in a partial filament eruption (Tang 1986; Gilbert
Filaments, also called prominences when they appear above
et al. 2000; Gibson et al. 2002; Pevtsov 2002). For a detailed
the solar limb, are a cool plasma material that is suspended in
classification of filament eruptions, we refer to Gilbert et al.
the solar corona, supported and confined by the coronal
(2007). The full and partial eruptions are generally larger in
magnetic field. Filaments are closely related to the most violent
scale and more energetic, while the failed eruptions are often
solar activities, i.e., filament eruptions, solar flares, and coronal restricted to a local area; though, the association with GOES M
mass ejections (CMEs). In many cases, but not all, a filament is or X-class flares has also been reported in some cases.
identified to exist in the core of a CME (Hundhausen 1999). Solar eruptions are suggested to be initiated by ideal MHD
Thanks to high-resolution observations, a growing number of instabilities or, alternatively, by nonideal fast magnetic
recent studies suggest that filaments sometimes consist of a reconnection. The torus instability and the helical kink
double-decker or even multistranded structure before erupting instability are two ideal MHD instabilities of a flux rope or
(Liu et al. 2012; Li et al. 2017; Awasthi et al. 2018; Su et al. current channel often reported as trigger mechanisms for
2018; Pan et al. 2021). If two filaments are associated with the eruptions. In a general definition (Priest 2014; Ch. 2.9.1), a
same magnetic polarity inversion line (PIL) and separated in magnetic flux rope is a twisted flux tube, with the boundary of
height, such a system is called a double-decker structure (Liu the flux tube being given by a set of field lines that intersect a
et al. 2012). The two filaments may erupt together or simple closed curve. Torus instability occurs when the
sequentially (e.g., Zhu et al. 2015), or in some cases, the overlying field of a filament attenuates fast enough with height
upper one erupts outward while the lower one stays almost (Kliem & Török 2006). The filament will then rise faster and
unchanged (Liu et al. 2012; Zhu & Alexander 2014; Zheng faster, and it may eventually erupt into the interplanetary space.
et al. 2019). The threshold for the onset of torus instability is defined in
Solar filament eruptions have been classified into three terms of the decay index of the external poloidal field at the
categories, i.e., full, partial, and failed eruptions. In a full position of the current channel, i.e., n = -¶ ln Bex /¶ ln h. The
eruption, the entire magnetic structure erupts with most of the canonical values of the critical decay index are 1.5 and 1.0 for
preeruptive filament mass (Tang 1986; Plunkett et al. 2000). toroidal and straight current channels, respectively (Bate-
While in a failed filament eruption, the filament initially shows man 1978; van Tend & Kuperus 1978). In theoretical
an eruptive-like rise, and then falls back toward the Sun after calculations and numerical simulations by Kliem & Török
reaching a maximum height (e.g., Ji et al. 2003; Alexander (2006), Fan & Gibson (2007), Aulanier et al. (2010), and

1
The Astrophysical Journal, 923:142 (16pp), 2021 December 20 Chen et al.

Démoulin & Aulanier (2010), the critical decay index has been 3. Observations
found to lie between 1.0 and 2.0. The helical kink instability 3.1. Activities Prior to the Eruption
will occur if the magnetic field supporting the filament
accumulates a sufficiently strong twist, e.g., Φcr = 2.5π for a The GOES light curve of the event is presented in
uniformly twisted, line-tied flux rope (Hood & Priest 1979) and Figure 1(a). Between about 4 and 1.5 hr before the eruption,
Φcr ∼ 3.5π for more realistic flux rope models as expected on filament oscillations occur then decay (Figure 1(b)). This is
the Sun (Fan & Gibson 2003, 2004; Török et al. 2004). followed by the C1.3-class precursor flare. Both the filament
Exceeding this threshold, the flux rope will erupt, and part of its oscillations and the precursor flare are closely related to the
twist will transfer into writhe. The instability then saturates intermittent, bright ejections from the northwestern footpoint of
the lower filament, and the ejections are associated with flux
quickly; however, when it lifts the rope above the critical
convergence and cancellation in the photosphere. These series
height for the onset of the torus instability, a full eruption can
of preflare activities are presented in Figures 1–2 and described
result. in detail in the following.
In a confined or failed eruption, the filament first rises, then
halts, and finally falls back toward the Sun (e.g., Ji et al. 2003).
3.1.1. Intermittent Bright Ejections
Several mechanisms can explain the confinement of the
eruption; see Hassanin & Kliem (2016) for a detailed review. Prior to the eruption, both filament segments in the double-
The two most frequently inferred mechanisms are the decker structure exhibit a C shape, aligned along the same PIL
following. In eruption models based on ideal MHD instability in the outer (northeastern) part of the active region, and largely
(van Tend & Kuperus 1978), the eruption fails when the overlap each other (Figures 1(c)–(d)). However, the north-
condition for the torus instability is not met in a height range at western footpoints of the two filaments show a clear separation.
some distance above the initial destabilization of the flux (e.g., The location and shape of the upper filament change
Török & Kliem 2005; Guo et al. 2010). On the other hand, a dramatically during the eruption, as shown in Figures 1(e)–
(f). It is not possible to determine how much the filaments
failed filament eruption can also occur due to the magnetic
separate in height before and after the eruption, because data
reconnection between the erupting and overlying flux, which is
from a side view are not available. However, there is evidence
facilitated by the existence of a coronal magnetic null point suggesting the independence of the two filaments already
spanning a dome-shaped magnetic fan surface above the before the eruption, in addition to the separation of their
eruption site (e.g., Wang & Liu 2012). northwestern footpoints. Bright material is repeatedly ejected
In this work, we study the failed eruption of a filament along the lower filament from its northwestern footpoint toward
system with double-decker structure. The activities which the northeast, selected examples of which are marked with
occurred near the northwestern footpoint of the lower filament white arrows in Figures 1(g)–(j). While the lower filament is
are found to be crucial in the evolution of the system imbalance disturbed by such intermittent ejections, the upper filament
and the eruption of the upper filament. We analyze the remains stable until the onset of the eruption, suggesting that
morphology and evolution of this filament system in detail, in the upper filament is sufficiently separated from the lower one,
order to figure out what initiates the eruption and why the i.e., a double-decker configuration. Further evidence is
eruption fails. This paper is organized as follows. Section 2 provided by signatures of reconnection in the space between
presents the data set and instruments. Observational analysis the filaments at the onset of the eruption (see Section 3.2).
and results are presented in Section 3. Magnetic modeling of
the double-decker system and evolution of the eruption are 3.1.2. Oscillations and Slow Displacement of the Upper Filament
presented in Section 4. We discuss the initiation and confining The upper filament begins to oscillate longitudinally after the
mechanisms of the failed eruption in Section 5. A summary and bright ejection at 18:30 UT, which might have triggered the
the conclusions are given in Section 6. oscillation. The time–distance plot in Figure 1(b) shows that
the oscillation lasts at least five periods, and the oscillation
periods are not uniform, ranging from 30 to 40 minutes. The
2. Data Set and Instruments nonuniformity of the oscillations may be due to disturbances by
A failed eruption of a double-decker filament system occurs the intermittent ejections. As soon as the oscillations totally
in NOAA AR 12104 on 2014 July 5. This eruption is disappear at about 21:10 UT (about an hour before the C1.3
associated with a C2.5-class solar flare, but does not produce a precursor flare), a slow westward displacement of the upper
filament becomes apparent, suggesting the onset of a slow-rise
CME. The Solar Dynamics Observatory (SDO) satellite
phase. The projected velocity is estimated from Figure 1(b) to
(Pesnell et al. 2012) observes the whole eruption. EUV and
be ≈37 km s−1.
magnetic field observations are provided, respectively, by the
Atmospheric Imaging Assembly (AIA; Lemen et al. 2012) and
3.1.3. Precursor Flare
the Helioseismic and Magnetic Imager (HMI; Schou et al.
2012) on board SDO. The AIA images have a temporal After the filament oscillation, intermittent bright ejections are
resolution of 12 s, and a pixel size of 0″.6. The HMI line of still ongoing along the lower filament. A particularly strong
sight (LOS) and vector magnetograms have a temporal ejection around 22:05 UT from the northwestern footpoint of
resolution of 45 s and 720 s, respectively, and their pixel size the lower filament coincides with the onset of the C1.3
is 0″.5. The Hα images are provided by the Global Oscillation precursor flare, which peaks at 22:20 UT, as shown in
Network Group (GONG). The spatial and temporal resolutions Figures 1(k)–(n). The flare begins with the appearance of two
are about 1″ per pixel and 1 minute, respectively. bright flare ribbons (R1 and R2), located on the two sides of the
lower filament, at 22:12 UT. The lower filament does not

2
The Astrophysical Journal, 923:142 (16pp), 2021 December 20 Chen et al.

Figure 1. Overview of the activities prior to the main flare. (a) GOES 1–8Å soft X-ray light curve from 19:00 UT to 23:59 UT. The gray dashed lines refer to the start
(19:10 UT) and end (21:08 UT) time of the filament oscillation. The black dashed and solid lines represent the peak time of the precursor and main flares. (b) Time–
distance plot of the filament oscillation. (c)–(f) Images of the double-decker filaments (white arrows) before and after the eruption in 304 Å (AIA) and Hα (GONG).
The white dashed line in panel (d) shows the position of the selected slit for the plot in panel (b). (g)–(j) Intermittent bright ejections (white arrows) from the northwest
footpoint of the lower filament toward the northeast, observed in 131 Å (AIA) before the eruption. (k)–(n) Evolution of the precursor flare observed in AIA 304 Å.

change significantly, whereas the upper filament continues its Figures 1(i)–(n)). This provides strong evidence that the upper
slow westward displacement. The bright ejection clearly moves filament is embedded in twisted flux. This could be a flux rope
under the upper filament to the outer side of the filament or a flux rope in formation, not a magnetic arcade. The
channel and largely parallel to the lower filament in this observed upward turn of the ejected material can only be
northern leg of the C-shaped structure. After reaching the outer enforced by the magnetic field. Due to the very small values of
side of the filament channel, the bright ejected material rises the plasma beta at a low height in active regions (Gary 2001),
and wraps around the upper filament (white arrows in moving plasma must usually follow the existing field direction,

3
The Astrophysical Journal, 923:142 (16pp), 2021 December 20 Chen et al.

Figure 2. Left panel: AIA 171 Å image overlaid with red (positive) and black (negative) contours representing the LOS photospheric magnetic fields observed by
HMI. Right panel: temporal evolution of the total unsigned LOS magnetic field strength (red) and the average horizontal magnetic field strength (blue) within the
region enclosed by the white box in the left panel. The pink lines mark the peak times of the precursor flare (22:20 UT) and the main flare (22:43 UT), respectively.
The gray dashed lines refer to the time of the intermittent ejections shown in the third row of Figure 1. An animation of the AIA 171 Å images overlaid with red
(positive) and black (negative) contours representing the LOS photospheric magnetic fields observed by HMI with the same field of view as the left panel, covering the
time interval from 18:29 UT on July 5 to 05:58 UT on July 6, is available. The real time duration of the animation is 15 s.
(An animation of this figure is available.)

which here reveals left-handed twist. From its position above magnetic flux decay. An increase of the horizontal magnetic
the upper filament, the ejected material immediately begins to field component during or after eruptions has been reported,
drain toward the central region surrounded by the filaments, e.g., by Wang et al. (2012) and Liu et al. (2018a), and
which suggests that the flux rope embedding the upper filament suggested to be a back-reaction of the photosphere and inner
is not yet fully coherent. Associated with the curved ejection is part of the Sun to the flares or CMEs. Alternatively, Barczynski
the brightening of another ribbon (R3 in Figure 1(n)). The et al. (2019) argued that photospheric horizontal fields are
ribbon R3 forms at the periphery of a simultaneously enhanced by the reconnection-driven contraction of sheared
developing dimming area, which it fully encloses from flare loops in the corona.
22:17 UT (Figure 1(n)). The dimming suggests that an
expanding flux rope is rooted in this area. This is also
supported by the draining of the ejected material toward this 3.2. Failed Filament Eruption and Main Flare
ribbon: the ejected material appears to be guided by the outer
layers of the magnetic flux holding the upper filament. Immediately after the onset of the precursor flare, the bright
ejected material wraps around the upper filament, and then
drains toward the central region surrounded by the filaments
3.1.4. Evolution of the Photospheric Magnetic Field
(see bottom row of Figure 1). Simultaneously (at ∼22:15 UT)
the rise of the upper filament accelerates, associated with a
To understand the cause of these activities, we examine the C2.5 flare that starts during the decay phase of the precursor
evolution of the photospheric magnetic field observed by and peaks at 22:43 UT; see Figure 3 and the associated
SDO/HMI as shown in Figure 2 and the associated animation. animation. Again, the lower filament remains nearly
The left panel of Figure 2 shows an 171 Å image taken by AIA unchanged. By ∼22:20 UT, several brightened threads appear
overlaid with red and black contours representing positive and along the northern leg of the C-shaped structure, and then wrap
negative polarity in the magnetogram, respectively. The right about the erupting upper filament in a left-handed sense similar
panel of Figure 2 presents the temporal evolution of the total to the bright ejection, and also pass under the filament, which
unsigned magnetic field strength (red) and the averaged excludes the possibility that they outline flux overlying the
horizontal magnetic field (blue) within the region enclosed by filament channel. Their rapid brightening must result from
the white box in the left panel of Figure 2. The pink dashed and heating by reconnection, which most likely involves ambient
solid lines mark the peak times of the precursor flare (22:20 flux in the space between the filaments in the standard manner
UT) and main flare (22:43 UT), respectively. The gray dashed of a two-ribbon flare. The energy deposition into the two
lines refer to the time of each bright ejection identified in the ribbons next to the lower filament also indicates such
EUV channels and shown in Figures 1(g)–(j). Prior to the reconnection, which additionally implies that the two filaments
eruption, two small magnetic polarities are observed to reside in separate magnetic structures already by the onset of
gradually converge and cancel each other at the source region the eruption, at least along the northern leg of the C-shaped
of the bright ejections as enclosed by the white box. The total filament channel. The threads show that the flux added to the
unsigned magnetic flux within this source region first increases upper filament by the reconnection carries twist. When it starts
then begins to decrease about 10 minutes before the precursor to rise, the filament has a straight and slim C shape with only
flare. These changes are due to flux convergence into the minimal indications of twist. However, it rolls in a way
selected area and subsequent flux cancellation. The horizontal resembling a twisted ribbon (∼22:30 UT), consisting of at least
component of the magnetic field within this source region two major intertwined threads (∼22:35 UT; white arrow in
increases rapidly after the peak of the precursor flare, and then Figure 3(f)), revealing that the upper filament is also internally
gradually decreases as the main flare and the total unsigned twisted.

4
The Astrophysical Journal, 923:142 (16pp), 2021 December 20 Chen et al.

Figure 3. Twisted internal structure of the erupting filament and release of twist during the main flare. Four snapshots in 304, 335, and 94 Å taken by AIA are
presented in the top, middle, and bottom rows, respectively. First column: the eruption begins and the upper filament starts to rise. Second column: the erupting upper
filament nearly reaches its maximum height. A clearly twisted structure (white arrow) is best seen in AIA 335 Å. Third column: the erupted filament untwists during
its draining from the terminal position at the west side of the C-shaped filament channel. Fourth column: after the eruption, the upper filament evolves into a much less
twisted structure. The C-shaped (outer) and inner ribbons of the main flare are marked as R4 and R5, respectively, and the corresponding flare loops are marked with
pink arrows in the bottom row. An animation of the AIA 304, 335, and 94 Å images with the same field of view as the first column of this figure, covering the time
interval from 22:05 UT to 23:00 UT, is available. The real time duration of the animation is 4 s.
(An animation of this figure is available.)

From a point early in the rise (∼22:20 UT), the erupting C-shaped ribbon (i.e., outer ribbon R4) enclosing the filament
filament also shows a motion in the direction to its footpoints. system; see Figures 3(c)–(d). The flare loops (pink arrows in
This rolling motion stops at 22:50 UT after the filament has Figures 3(h) and (l)) rooted in this ribbon connect to a small
acquired a reverse-C shape, and subsequently filament material cusp-shaped ribbon (R5 in Figure 3). Similar to ribbon R3 of
drains toward the Sun. Many brightenings appear intermittently the precursor flare, ribbon R5 is part of the strongly variable
within and below the erupting filament during this final stage of brightening that develops in the periphery of the dimming area.
the eruption (Figures 3(c) and (g)). The filament threads Thus, both ribbons form in the footpoint area of the erupting
become more or less straightened, indicating that the filament upper filament. The flare loops rooted in R5 additionally imply
loses essentially all of its twist during the draining that the erupting flux of the upper filament reconnects. The
(Figure 3(d)). This implies that the erupted flux reconnects reconnection involves ambient flux rooted in Ribbon R4.
with ambient flux, different from the standard 2D flare model.
Such reconnection is also suggested by the fact that the 4. Modeling and Results
draining to the negative-polarity side proceeds toward two To understand the morphology and evolution of the double-
separate footpoint areas: mostly toward the same region as decker filament system, we model the magnetic field using the
during the precursor and also toward the southwestern end flux rope insertion method developed by van Ballegooijen
region of the filament channel. When the upper filament is (2004). This method is proved to be capable of reproducing
close to its maximum height, a flare ribbon at the eastern outer nonlinear force-free fields before eruption, as well as unstable
side of the filament channel begins to brighten at 22:37 UT. magnetic configurations during the early phase of a successful
The ribbon then gradually extends and evolves into a full eruption, allowing the stability of filaments to be analyzed (e.g.,

5
The Astrophysical Journal, 923:142 (16pp), 2021 December 20 Chen et al.

a complex three-dimensional structure (Demoulin et al. 1996).

The intersection of two QSLs, a generalization of a two-
dimensional magnetic X structure, is referred to as a hyperbolic
flux tube (HFT). We investigate magnetic connectivities by
tracing field lines point-wise with a fourth-order Runge–Kutta
method, and use footpoint positions of field lines in the
photosphere to calculate Q. The results of the potential field
modeling and their comparison with SDO/AIA observations
are shown in Figure 5. We find that the overlying field of the
filament system is similar to the dome-plate geometry presented
in Chen et al. (2020). This is different from the widely accepted
model for circular-ribbon flares (Masson et al. 2009; Wang &
Liu 2012), since no null point is identified around the expected
location, but the large-scale structure of the field possesses
considerable analogy. Underlying this analogy is the fact that
the C-shaped filament channel encloses an area of (negative)
polarity and is nearly completely surrounded by the opposite
(positive) polarity (Figure 4). Figures 5(d)–(f) show the Q
factor isosurfaces of the potential field, computed using the
method by Liu et al. (2016). The magnetic configuration
appears to be in agreement with the conjecture by Wang et al.
(2014) that a quasi separator separates the dome in two parts.
Figure 4. Longitude–latitude map of the radial component of photospheric
magnetic field in the HIRES region of the model taken by SDO/HMI at 22:00
UT. The blue lines ending in circles show the selected filament paths for 4.2. Nonpotential Field Model
inserting the flux ropes. The yellow line (start location marked as a red asterisk)
refers to the location of the vertical slices in Figures 6–8. Positive and negative We have constructed 30 models in total, and then checked
polarity are shown in white and black, respectively. The white box marks the the relaxed models against the observations, in order to
approximate field of view (FOV) of Figures 5(a)–(b), the FOV of the left two
columns of Figure 6, and the FOV of Figure 8(h). The left and right magenta
constrain the axial (Φaxi) and poloidal fluxes (Fpol) of the
asterisks mark the positions of the solid and dashed curves representing the inserted flux bundles. The relaxation is performed with a
decay index in Figure 10. similar setting to Relaxation 1 as shown in Table 2 of Su et al.
(2011). The range of the parameter variations in this grid of
Su et al. 2011, 2015; Li et al. 2017; Liu et al. 2018d). A models is the following: Φaxi = [2, 6] × 1020 Mx; Fpol = [ − 1,
detailed description of the methodology and its application can 1] × 1010 Mx cm−1. Based on visual inspection, the best-fit
be found in the literature van Ballegooijen (2004), Su (2019), model is determined by comparing the selected model field
and references therein. First, a potential field is computed from lines with the observed filament before, during, and after the
the high-resolution (HIRES) HMI magnetogram embedded in a eruption as shown in the first column of Figure 6. Some
low-resolution HMI global synoptic map. The HIRES region as parameters, e.g., footpoint location and length of the inserted
shown in Figure 4 spans about 29° in both longitude and flux bundle, are well constrained by observations. For other
latitude, and the spatial resolution in the low corona is 0.001 parameters, there is a moderate freedom, e.g., for the most
Re. The resolution of the global field is about 1°. Both HIRES important parameter Φaxi, we find a range Φaxi = [3, 4] × 1020
and GLOBAL regions extend from the solar surface (r = Re) Mx to be consistent with the observations. The values
up to a source surface (r ∼1.7 Re) above which the magnetic Φaxi = 3 × 1020 Mx (lower flux bundle) and 4 × 1020 Mx
field is assumed to be radial. Then we create cavities and insert (upper flux bundle) yield the overall best agreement with the
two flux bundles with a double-decker structure along the data. Yet other parameters are not well constrained by the data;
observed filament paths. The resulting configuration is not in a in particular, the relaxed configurations are similar in the whole
force-free equilibrium, so the last step is to allow the field to considered range of Fpol.
relax through magnetofrictional relaxation (Yang et al. 1986;
van Ballegooijen et al. 2000). For an unstable model, like the 4.2.1. Confined Partial Eruption
one considered below, relaxation and instability compete in the
evolution. Eventually, the instability wins out, which is Figures 6–8 present results of the best-fit model with two
manifested as a gradual rise of the unstable flux. The flux bundles inserted along the blue paths shown in Figure 4.
magnetofrictional modeling also captures key topological The flux bundle FR1 connecting the footpoints L1 and L2
changes but not in the real timescale of the evolution (e.g., Su corresponds to the lower filament, above which is the shorter
et al. 2011; Janvier et al. 2016). flux bundle FR2 (connecting the footpoints U1 and U2)
corresponding to the upper filament, whose footpoint in the
negative polarity is chosen to lie near the area of draining
4.1. Potential Field Model
filament material. The initial poloidal fluxes of the inserted flux
We investigate the structure of the potential field by bundles are set to be 0 Mx cm−1, i.e., both flux bundles
computing the squashing factor Q, which is a measure of the initially contain no twist, to minimize the influence of the
gradient in field line mapping on a surface or between two insertion procedure on the resulting twist. The axial fluxes of
surfaces (Titov et al. 2002). Quasi-separatrix layers (QSLs), FR1 and FR2 are 3 × 1020 Mx and 4 × 1020 Mx, respectively.
defined by high Q values, are known to be the preferred The first column of Figure 6 presents three AIA images in
locations for the occurrence of reconnection, and often exhibit 304 Å before, during, and after the failed filament eruption.

6
The Astrophysical Journal, 923:142 (16pp), 2021 December 20 Chen et al.

Figure 5. Results from the potential field model. (a) Selected field lines from the potential field model overlaid on the AIA image in the 304 Å channel taken at 22:02
UT. The same field lines viewed from anther angle are shown in (b). The red and green contours represent the positive and negative photospheric magnetic field at
22:00 UT, respectively. (c) Photospheric log Q map for the potential field model. (d)–(f) Different views of three-dimensional high-Q (log Q = 4 ) isosurfaces. The
field of view of the X–Y plane is similar to that in panel (c), and the Z-axis unit is the cell size of our model (1 cell ∼ 0.7 Mm).

From the second to the fourth column, the individually selected surrounding fields, which is enabled by numerical diffusion
field lines and the distribution of the current density in the and considered in more detail below.
model are displayed after different numbers of magnetofric-
tional iterations. A comparison of the images in the first and 4.2.2. Magnetic Topology versus Observations
second columns shows that the selected field lines after 5000,
30,000, and 90,000 iterations do qualitatively match the The top two rows of Figure 7 show the evolution of log Q in
observed filaments before, during, and after the eruption. The the photosphere and in the vertical cross section along the
side view of the same field lines in the third column and the yellow line marked in panel (a). The platelike shape extending
current density distribution and field vectors in the cross section from the top of the dome-like structure (Figure 5) is the analog
in the forth column show that the upper flux bundle FR2 rises of the outer spine in a classical null-point topology. It is seen as
to near the top of the separatrix dome from 5000 to 30,000 the nearly straight high-Q line marked by green arrows in
iterations, showing an eruption as observed. At the same time, Figures 7(d)–(f). The log Q maps show significant changes
the upper flux bundle acquires twist. The resulting state during the relaxation process. The dome-like structure, outlined
conforms to the projected position and the twisted appearance as a high-Q arch in the cross section, first expands and rises,
and then shrinks. Simultaneously, the plate gradually moves
of the erupted upper filament (Figure 3(f)). Subsequently, the
toward the open side of the C-shaped structure, in a similar
twisted flux bundle stops its rise under the separatrix dome and
direction as the erupting filament.
is largely eroded (see fourth column), which will be analyzed in
After 90,000 iterations, the log Q map at a height of 2 Mm
detail in Section 4.2.3. The remaining flux under the top of the above the solar surface and the comparison with an image
dome shows an untwisted flux bundle with a slight inverse C observed in 304 Å are presented in the bottom row of Figure 7.
shape (second column), also largely similar to the observed Their comparison shows that the eastern outer ribbon of the
draining filament. The lower flux bundle remains nearly main flare, R4, compares favorably with the eastern arc of high-
unchanged throughout the relaxation, which also agrees with Q lines in the log Q map (marked by a red arrow).
the observations. After 90,000 iterations, it is only slightly
shorter and lower than at 5000 iterations. After 5000, 30,000,
and 90,000 iterations, the free magnetic energy in the best-fit 4.2.3. Twist Evolution
model is 1.08 × 1031 erg, 6.16 × 1030 erg, and 3.54 × 1030 erg, A flux rope is a group of helical field lines collectively
respectively, as listed in the fourth column of Figure 6. winding around a common axis (Priest 2014; Liu 2020), which
Therefore, the evolution of the observed filament eruption is called the magnetic axis. The winding is measured by the
corresponds to the decrease of free magnetic energy. This twist angle of the field lines about the axis, which can be
results from the decrease of the current flowing through the expressed through the local axial and azimuthal field
upper flux rope as the rope rises and likely also from magnetic components Bz(r) and Bθ(r), the length of the axis, l, and the
reconnection between the inserted flux bundles and the distance from the axis, r, as f(r) = lBθ/(rBz), if the flux rope is

7
The Astrophysical Journal, 923:142 (16pp), 2021 December 20 Chen et al.

Figure 6. The first column shows three AIA images in 304 Å before, during, and after the failed filament eruption. Top and side views of selected field lines from the
best-fit model after 5000, 30,000, and 90,000 relaxation iterations are presented in the second and third columns. The field lines were selected for each iteration
individually. The fourth column displays the corresponding distribution of the current density (gray-scale image) and magnetic vectors (black and white) in a vertical
slice at the position of the yellow line in Figure 4. The intersections of the field lines with the plane of the slice are marked with white dots. The red and green contours
represent the positive and negative polarities observed by HMI at 22:00 UT, respectively. The axis unit in the right column is the cell size of our model.

cylindrically symmetric. Often, a radial average, f, is referred As mentioned above, no twist (i.e., no poloidal flux) is
to as the twist of the flux rope. The twist number is N = f/(2π). carried by either of the inserted flux bundles initially. After
In practice, flux ropes are often complex structures, so it is 5000 iterations, i.e., at the moment before the eruption, twist
difficult to precisely locate the axis and perform the radial has accumulated significantly in the upper flux bundle, as
average. It is far more straightforward to compute a local twist shown in Figures 8(a) and (d). The local twist in the upper flux
number of each individual field line bundle Tw<−1 (blue area; blue arrows) has its sign in
Tw = ∫( ∇ ×B) · B/(4πB2)dl, where dl is the line element along agreement with the prevailing helicity in the filament channel
a field line (Berger & Prior 2006). This quantity measures how as revealed by the erupted filament and the brightenings
the field lines in an infinitesimally small vicinity wind about the preceding the eruption. The area of highest |Tw| is close to a
considered field line, i.e., it does not refer to the axis of the flux swirl in the in-plane field vectors in Figure 8(g), so the
rope. In the vicinity of the axis, Tw is close to N, and agrees formation of a flux rope is indicated, with Tw here largely
with N if the rope is axially symmetric. As an indication, representing true twist about the axis in the center of the swirl.
The maximum local twist of the upper flux rope reaches
moderate differences of N being smaller than Tw by ∼0.5–1
Tw ≈ −1.75 (roughly corresponding to f ≈ −3.5π). This is
have been found in two specific solar cases (Liu et al. 2016;
close to the threshold of the helical kink instability (Fan &
Kliem et al. 2021). If there exists a flux rope, a map of Tw Gibson 2003; Török et al. 2004). However, the average local
should generally reveal its existence and location, and provide twist (Tw ≈ −1.45) remains clearly lower and may be even
a rough estimate of its twist f. However, one has to be careful lower if N < Tw. Therefore, the modeling suggests that the
in the interpretation of such maps, because Tw is nonzero also helical kink instability did not commence, in agreement with
in current layers, which complicates the interpretation. We map the indication from the observations that lack clear signatures
Tw by integrating its local density, α/4π, where α = μ0J∥/B is of the development of writhe.
the force-free parameter, along each field line, using the code The local twist accumulated by the lower flux bundle is far
package provided by Liu et al. (2016). The calculated Tw maps weaker and positive (0 < Tw < 1; red area and green arrows).
after 5000, 30,000, and 90,000 iterations are presented in the The in-plane field vectors in Figure 8(g) do not show any
left, middle, and right columns of Figure 8, respectively. The indication of swirl in this area, so the local twist Tw here does
green and pink contours indicate the areas with local twist not indicate any true twist about a magnetic axis. Overall, we
number |Tw| > 1 and |Tw| > 1.75, respectively. find that the filament channel contains a flux rope atop a flux

8
The Astrophysical Journal, 923:142 (16pp), 2021 December 20 Chen et al.

Figure 7. Maps of log Q for the best-fit model and comparison with observations. log Q maps at a height of z = 1.8 Mm (top row) and vertical slices of log Q map
(middle row) after different magnetofrictional iteration steps along the yellow line with the start point marked as a black asterisk, which is the same as the yellow line
in Figure 4. The axis unit in the middle row is the cell size of our model. Bottom row: log Q map for the model after 90,000 iterations of relaxation at a height of 2
Mm above the solar surface (g) and the same log Q map (blue) superimposed in the image observed in 304 Å by AIA (h). Green arrows in the middle row mark the
plate-shaped QSL. The red arrow in the bottom row marks the high-Q area on the eastern side of the inserted flux ropes.

bundle with negligible twist. This bears some similarity to indicate a flux rope. However, there continues to be no support
another active region analyzed in Awasthi et al. (2019), whose for the presence of a flux rope structure from the map of the in-
structure was confirmed by mass motions. plane field vectors (Figure 8(h)). The elongated shape and
After 30,000 iterations, the area with Tw < −1.75 has built location of this relatively high-Tw area are instead suggestive of
up clearly in the upper flux rope (Figures 8(b) and (e)) and a current layer steepened between the dome and the rising
continues to enclose the magnetic axis (Figure 8(h)). The upper flux bundle. Overall, the stable lower filament is modeled
increase of the twist is also clearly seen in the field line plots in by the flux bundle of negligible twist that stays at its original
Figure 6. In contrast, the local twist number stays at low height, and the erupting upper filament is modeled by a rising
positive values, Tw < 1, in the original area of the lower flux rope of increasing twist.
inserted flux bundle. The field lines started in this area continue A comparison of the middle and right columns of Figure 8
to show a flux bundle of negligible twist that runs along the shows a significant decrease of the twist in the upper flux
whole C-shaped filament channel (Figure 6; middle row). bundle during the further evolution from 30,000 to 90,000
Additionally, an elongated region of positive Tw appears iterations. At the same time, the twist number increases
between the upper flux bundle and the dome, and acquires a significantly in the flare ribbons R4 and R5 of the main
higher local twist of Tw ∼ 1, which is often considered to eruption (black arrows). This evolution suggests that the twist

9
The Astrophysical Journal, 923:142 (16pp), 2021 December 20 Chen et al.

Figure 8. Evolution of local magnetic twist Tw in the nonpotential model. (a)–(c) Photospheric Tw maps after different iterations of the magnetofrictional relaxation.
(d)–(f) Corresponding vertical Tw maps along the yellow line marked in Figure 4. Green contour: ∣ Tw ∣ = 1.0;. Pink contour: ∣ Tw ∣ = 1.75. (g)–(i) Vertical maps of the
current density in the same vertical slice overlaid with in-plane field vectors. The approximate axis positions of the upper and lower flux bundle are marked with blue
and green arrows, respectively. The critical height for the onset of the torus instability is shown by a magenta asterisk, in (g)–(h). The axis unit in the bottom two rows
is the cell size of our model.

of the upper flux rope is transferred to the post-flare arcade field diffuses away around the lower bundle in the first few thousand
whose footpoints are highlighted by these flare ribbons. In iterations and becomes less well defined in maps of the current
these locations, the increase in the local twist Tw indicates the density compared to Tw maps. This results from the present
steepening of a current layer (a QSL according to Figure 7(c)), model’s tendency to develop its current density in the volume
rather than the buildup of another flux rope. of the inserted flux bundles, not only in the surface layer. The
The buildup of twist in the upper flux bundle is a natural Tw map becomes the most sensitive diagnostic of the annulus,
result of the magnetofrictional evolution after the insertion of which is still well defined for the flux rope at 5000 iterations
the flux bundle along the PIL. The ambient potential field and still indicated at 30,000 iterations (Figures 8(d)–(e)). This
points essentially perpendicularly to the PIL and the flux magnetofrictional buildup of the current density and twist in the
bundle. The magnetofrictional relaxation is a strongly diffusive flux rope proceeds at scales of several 104 iteration steps, i.e.,
process that creates a transition layer of smoothly changing continues into the rise phase of the rope (Figure 6).
field directions between these flux systems, with the current In the magnetofrictional evolutionary model, the twist of the
(which represents the free energy) mainly flowing in this layer. flux rope is built up in place by diffusion of the magnetic
The field directions in the transition layer are intermediate information from the inserted and the ambient flux, into the
between the axial and perpendicular directions, therefore layer between them. There is no bodily transport of twisted flux
yielding helical field lines that wrap around the inserted flux into the layer. This differs from the relevant path of MHD
bundle, which becomes a flux rope. evolution of a nearly ideal magnetofluid, which represents the
Initially, the layer forms as an annulus around each of the conditions in the corona. In this case, twist can be transported
inserted flux bundles and can be seen in cross-sectional maps of along a flux rope or added at the surface as new layers of
the current density, Q, and Tw. However, the layer largely twisted flux. The former would occur if the flux rope is in the

10
The Astrophysical Journal, 923:142 (16pp), 2021 December 20 Chen et al.

Table 1 these field lines of the flux rope have changed into field lines
Global Nonpotential Measures after Different Iterative Relaxations representing flare loops that connect the main flare ribbons
Iteration Efree/Epoten Hr/F2u 〈Tw〉box 〈Tw〉FR
(pink arrows in Figure 3). This evolution demonstrates
(10−3) (10−4) (10−2) (10−2) magnetic reconnection between the rising flux rope and
ambient arcade-like flux that is rooted in the outer ribbon R4.
5000 9.77 −3.89 7.39 −24.87
The reconnecting flux rope field lines obtain a new footpoint in
30,000 5.60 −3.86 5.29 −20.49
90,000 3.22 −3.32 3.84 −2.32
the large ribbon R4 and produce flare loops, as termed ar-rf
reconnection by Aulanier & Dudík (2019). The upper flux rope
Note. Global nonpotential measures include magnetic free energy (EFree) begins to lose twist when this reconnection occurs near the
divided by the potential field energy, flux-normalized relative magnetic helicity intersection of the dome- and plate-shaped QSLs. This transfer
(Hr/F2u), average local twist of the HIRES region (〈Tw〉box), and average twist of twist from the flux rope to the flare loops (twist release) is
of the flux rope of the best-fit model (〈Tw〉FR) after different iterative clearly seen in the twist maps in Figures 9(c) and (f) as a
relaxations. The potential field energy is 1.1 × 1033 erg. decrease of Tw at the footpoint areas of the flux rope and a
simultaneous increase of Tw at the footpoint areas of the
reconnected field lines. In addition to this transfer of twist,
process of emergence (Parker 1974) or if a vortex would act at there is a reduction of the average twist in the box, due to the
its footpoint(s); however, neither of these is observed in the numerical diffusion of the magnetofrictional scheme (Table 1).
course of the present event. The latter results from the standard To compare the transfer of twist by reconnection with the
flare reconnection underneath an erupting flux rope and, in the loss of twist by numerical diffusion, we compute the average
present event, is not only expected but also indicated by the degree of local twisting in the remote footpoint area of the flux
bright threads wrapping around the upper filament during its rope, á Tw ñFR , at i = 30,000, marked by a green box in
rise. Despite the difference in the way the twist is built up, the Figure 9(c), and its change by i = 90,000. Different from the
underlying reason is the same, i.e., the conversion of mutual- flux in the footpoint area marked with the asterisk in Figure 9,
helicity into self-helicity during the interaction of the flux rope the flux rooted in the remote footpoint area does not connect to
with the ambient field. the places that the twist is transferred to, therefore providing a
Table 1 lists various quantities of the best-fit model after measure of the twist released from the flux rope. The integral of
different relaxation iterations, all computed in the HIRES Tw(x, y) in the remote footpoint area is divided by the footpoint
region. The average degree of local twisting, á Tw ñbox , is area for the correct average. It is found that 89% of á Tw ñFR at
computed by integrating the absolute values of Tw over the i = 30,000 is released by i = 90,000. This is much larger than
bottom area of the HIRES volume and dividing this by twice the average decrease (27%) of á Tw ñbox by numerical diffusion in
the bottom area (because the local twist of each field line is the same period. This lends support to our conjecture that a
summed up twice, once at each footpoint). Tw at the bottom magnetofrictional description can capture some essential
plane is the value integrated over each individual field line structural changes of a magnetohydrodynamic evolution, such
rooted in the bottom plane, i.e., Equation (7) in Liu et al. as those changes due to magnetic reconnection. However,
(2016). Therefore, the change of Tw in the bottom plane comparative studies of magnetofrictional versus MHD model-
represents the change of magnetic twist in the relevant three- ing are needed to obtain a deeper understanding of their
dimensional volume. Berger & Field (1984) established the respective capabilities.
relative magnetic helicity, Hr, as a relevant physical quantity in
magnetically open volumes, like the solar corona. Hr measures 4.2.4. Effect of Evolving Magnetogram
the linkage of magnetic flux tubes in the system, and is
calculated following Appendix B in Bobra et al. (2008). In our simulation, the normal component of the magnetic
Table 1 shows that all of the global nonpotential measures, field in the bottom plane is fixed during the magnetofrictional
including the free magnetic energy divided by the potential evolution, but the magnetic field in the solar surface does
field energy, the flux-normalized relative magnetic helicity, the change somewhat during the relatively long-lasting eruption (
average degree of local twisting of the HIRES region, and the i.e., from 22:02 UT to 22:52 UT). In order to investigate the
average twist of the flux rope, decrease weakly with increasing effect of the magnetic field change on the modeling results, we
iteration numbers. From i = 30,000 to i = 90,000, 27% of have also constructed both a potential and a nonpotential
á Tw ñbox and 14% of the flux-normalized relative magnetic magnetic field model based on the LOS photospheric
helicity, Hr Fu2 , where Fu is the unsigned magnetic flux magnetogram taken at 22:52 UT and performed the corresp-
through the bottom plane, have decayed by numerical onding topological analysis. We find that the values and shape
diffusion. of the high-Q and high-Tw volumes do change, but the key
In order to understand the role of magnetic reconnection in features remain solid. In particular, the structure of a flux rope
the twist release of the upper flux rope, we investigate the atop a flux bundle of negligible twist and the buildup and
change of the field lines in the model between 30,000 and release of twist of the erupting upper flux rope are essentially
90,000 relaxation iterations in Figure 9, using a procedure the same.
similar to that presented by Aulanier & Dudík (2019). At
30,000 iterations, we trace the field lines inside the erupted flux 5. Discussion
rope from a vertical slice to their photospheric footpoints
5.1. Initiation of Eruption
(panels (a)–(c)). The field lines are then traced from the
footpoints in the negative polarity (red asterisk), which lie near Kliem et al. (2014) have discussed three possible trigger
the observed ribbon R5 shown in Figure 3, in the model after mechanisms for the eruption of the upper part in a double-
90,000 iterations (panels (d)–(f) of Figure 9). The comparison decker filament: (1) a weakening of the overlying field, or a
and the online animation associated with Figure 9 show that transfer of flux from the lower to the upper filament; (2) a

11
The Astrophysical Journal, 923:142 (16pp), 2021 December 20 Chen et al.

Figure 9. Change of the selected field lines with fixed southern footpoints (marked with a red asterisk) from 30,000 (top row) to 90,000 iterations (bottom row). The
left and middle columns show the field lines going through the vertical log Q and Tw maps of Figures 7 and 8. The bottom image in each panel shows the LOS
photospheric magnetogram taken at 22:00 UT by HMI. The right panel shows the selected field lines overlaid on the photospheric Tw maps from the observation view
angle. An animation of the field line plots with the same field of view as the third column of this figure, covering the time interval from i = 5000 to i = 90,000, is
available. The real time duration of the animation is 2 s.
(An animation of this figure is available.)

tether-cutting reconnection at the HFT between two filament phase, occurs before the above potential triggers. The only
flux ropes, which also transfers flux to the upper rope; and (3) associated process in the corona is the decaying filament
the buildup of supercritical twist for onset of the helical kink oscillation, which fails to trigger the eruption in the present
instability only in the upper flux rope. event. Therefore, the displacement may result from long-lasting
In our case, most of the intermittent bright ejections flow tether-cutting reconnection driven by photospheric flux can-
along the lower filament. Only the strongest one, which leads to cellation (Section 3.1.4).
the precursor flare, also shows an upward displacement, in In instability models for eruptions, a trigger may only push
addition to its motion along the lower filament, before the magnetic structure across the boundary of stability, but the
wrapping around the upper filament. This ejection may transfer subsequent onset of instability drives the eruption. Therefore,
some flux from the lower to the upper filament. Since the main we have to ask whether an ideal MHD instability could have
eruption evolves seamlessly out of this ejection, it is possible followed the potential triggers identified above. While the
that the ejection plays the role of the final drop, destabilizing a helical kink is not supported, our modeling does support the
structure on the verge of instability. occurrence of the torus instability.
The same triggering mechanism can be provided by the The canonical value of the critical decay index for the onset
reconnection of ambient flux between the filaments during the of the torus instability is ncr = 1.5 (Kliem & Török 2006),
C1.3 precursor flare, which is suggested by the occurrence of which is considered as a reference in our analysis. We find that
brightened threads wrapping around the upper filament and by the corresponding critical height, computed from the horizontal
the ribbon pair R1 and R2 marked in Figure 1(n) (Section 3.2). component of the potential field, varies significantly along the
Such reconnection is in line with the findings reported by Liu eruption trajectory. At the eruption onset, the estimated axis
et al. (2012) and Zhu et al. (2015). height of the upper flux rope (the center of the swirl in the
The maximum twist angle of the upper flux rope after 5000 magnetic vector distribution) of 21 Mm exceeds the critical
iterations in our model reaches ∼3.5π (provided N is not height for the torus instability (16 Mm; magenta asterisk); see
significantly smaller than Tw), which is close to the threshold of Figures 8(g) and 10. This suggests that the instability initiates
the helical kink instability. However, the average twist is lower, the upper filament eruption. Similarly, the magnetic flux rope
and the observations show no signature of twist-to-writhe axis in the best-fit model after 30,000 iterations has a height of
conversion, which is the typical telltale sign that the kink ∼41 Mm, slightly above the critical height of 36 Mm; see
instability is involved in the destabilization of a flux rope. Figures 8(h) and 10. This is fully consistent with the status in
Therefore, the occurrence of this instability in the event is the course of the eruption.
unlikely. As a further potential eruption mechanism, the double arc
One must keep in mind that the first compelling sign of instability (DAI) has been proposed as a precursor process or as
evolution toward eruption, i.e., the onset of a slow westward an alternative to the torus instability (Ishiguro & Kusano 2017;
displacement of the upper filament, suggestive of a slow-rise Kusano et al. 2020). This instability can result when tether-

12
The Astrophysical Journal, 923:142 (16pp), 2021 December 20 Chen et al.

Figure 10. Decay index height profiles (left panel) computed from the horizontal component of the potential field (right panel). The positions of the solid and dashed
curves relevant at i = 5000 and i = 30,000, respectively, are marked as magenta asterisks in Figures 8(g) and (h), as well as in Figure 4.

cutting reconnection merges two arc-shaped flux bundles into a Our analysis indicates that the confinement of the erupted
double-arc shape, which implies an additional upward Lorentz upper filament in the investigated event can be due to a
force in the bent flux at the reconnection point. The data of the combination of the aforementioned two main mechanisms.
event suggest the occurrence of topologically similar tether- Certainly the reconnection with overlying flux is important, as
cutting reconnection at the HFT between the two filaments demonstrated by the draining of the filament material in the
from the brightening of multiple sheared loops during the observations and as corroborated by the model.
precursor flare (see Section 3.1.3, Section 3.2, and the The observations and the potential field computation indicate
animation of the AIA 171 Å data in Figure 2). However, the that the tension force of the overlying flux is also important,
multiple loops represent a layer of flux wrapping around the providing a confining cage. The animation of the AIA data
upper filament, sharing its magnetic axis; they do not represent (Figure 3) shows that the first signs of the reconnection with
a new, double-arc-shaped flux bundle and current channel, ambient flux—the formation of new connections in the area of
which is the prerequisite equilibrium structure of the DAI. The the original negative footpoint—occur only after 22:39 UT. At
only clear flux rope structure suggested by the observations and this time, the filament has already nearly reached its final
the modeling is the upper filament, which maintains the
position. This indicates that the filament is strongly decelerated
structure of a single arc throughout the event. On the other
by the downward tension force of overlying flux, and the
hand, the Lorentz force in the upper reconnected flux
deceleration process is also in agreement with the large-scale
(visualized by the brightened multiple sheared loops) points
upward and supports the subsequent eruption, similar to the structure of the overlying flux as revealed by the potential field.
action of a DAI. Figure 10 shows that the flux above the QSL dome provides an
extended height range where n < 3/2. A rising flux rope can be
stopped in such a torus-stable height layer if the layer contains
5.2. Confinement a sufficient amount of flux (DeVore & Antiochos 2008). We
expect that this happened in the investigated event. The two
Various mechanisms can stop an eruption. Observations and flux systems pressing against each other are also likely to
simulations reveal that confined eruptions may occur because reconnect, and this is indeed indicated by the observations
the strong overlying magnetic loops form a confining cage that during the draining of the stopped filament. On the other hand,
can prevent the escape of flux ropes (Török & Kliem 2005; in the magnetofrictional evolution of the model flux rope,
DeVore & Antiochos 2008; Guo et al. 2010; Amari et al. 2018; reconnection with overlying flux is the dominant process that
Yang et al. 2019; Zheng et al. 2019). Such a strong strapping stops the rise, due to the high diffusion inherent in this
field above the eruption site provides a torus-stable regime
approximation.
above a torus-unstable one (see examples in Wang et al. 2017). The torus-stable height layer in the present active region has
In addition, a flux rope is unable to erupt successfully if the
also been found by Filippov (2020). Assuming that a strapping
tension force of a guide field is sufficiently strong (Kliem et al.
field of opposite direction above a magnetic null would exert an
2014; Myers et al. 2015). Alternatively, confined eruptions
often occur below a dome-shaped magnetic fan surface with a upward Lorentz force on the rising flux rope, this author
coronal magnetic null point. (e.g., Masson et al. 2009; Liu et al. concluded that “the most reasonable force that terminate the
2018c; Yang & Zhang 2018). A magnetic null facilitates the filament ascending above the null point is the gravity force.”
onset of magnetic reconnection because oppositely directed This would be valid in vacuum fields. However, in MHD, a
flux exists there in close proximity. In our case, the erupting flux rope rising into a region of oppositely directed strapping
filament is surrounded by a QSL of similar dome-plate field will never penetrate into this flux; thus, it will not feel
geometry (see the opposite field directions in Figure 5(b)). such a Lorentz force. Rather, the rope will either stretch this
When the rising upper filament meets the overlying flux at the flux upward, creating a downward magnetic tension force
dome region (Figures 7–8), magnetic reconnection between (DeVore & Antiochos 2008; Kliem et al. 2021), or will
them indeed occurs (Figure 9), which erodes the flux in the reconnect with this flux, which erodes the flux rope, weakening
rope. This changes the forces in the rope and can halt the its upward Lorentz force (Török & Kliem 2005; Hassanin &
eruption (e.g., Török & Kliem 2005; Hassanin & Kliem 2016). Kliem 2016).

13
The Astrophysical Journal, 923:142 (16pp), 2021 December 20 Chen et al.

6. Summary and Conclusions negligible twist, which appears to closely match the observed
double-decker filament at the eruption onset. Although both
We investigate the failed partial eruption of a C-shaped
flux bundles were inserted without an azimuthal field
double-decker filament system, which occurred in AR 12104 in
component, the upper one acquires twist. The twist continues
the southern solar hemisphere on 2014 July 5. The filaments
to increase until the termination of the rise after 30,000
enclose a region of negative flux that is nearly completely
iterations. Consistent with the observations, the twist in the
surrounded by positive flux. Before the eruption, a sequence of
upper flux rope shows a significant decrease subsequently. At
interrelated activities are observed: intermittent bright ejec-
the same time, the nonpotentiality indicated by the photo-
tions, oscillation of the upper filament, and a C1.3 precursor
spheric Tw map increases substantially in the area of the main
flare. The eruption of the upper filament is associated with the
flare ribbons, suggesting a twist or current transfer from the
main C2.5 flare with a C-shaped flare ribbon surrounding the
erupted flux rope to the surrounding field, which results from
stable lower filament and a compact ribbon forming in the
the reconnection of the flux rope with the surrounding magnetic
enclosed polarity. The upper filament displays a prominent
field at the high-Q dome. The ribbons take the form of an
twisted structure when it is undergoing a failed eruption and
extended C on the outer side of the filament channel and of a
rolls over toward the open side of the C shape during the
compact area inside, respectively. The C-shaped outer ribbon
eruption. The rise of the filament then halts; it transforms to a
coincides with lines of high squashing factor Q in the
reversed-C shape and subsequently drains, partly to places
photosphere, similar in shape to the footpoint area of the
different from its original footpoints.
magnetic dome in a null-point configuration. The compact
A careful examination of the photospheric magnetic field
inner ribbon forms at the periphery of the erupting flux rope of
suggests that the intermittent ejections are due to flux
the upper filament, as indicated by a developing dimming. This
cancellation between two small converging polarities near the
is located near the footpoint area of the magnetic spine.
northwestern footpoint of the filament. Long-lasting filament
Previously, it has been found that the magnetofrictional
oscillations are often observed before the onset of solar
relaxation process is capable of emulating the observed
eruptions; therefore, these oscillations were proposed as a
evolution of an eruption in the early phase (Su et al. 2011;
precursor leading to eruption (Chen et al. 2008). However, the
Savcheva et al. 2016; Liu et al. 2018d). For the first time, this
filament oscillation that lasts about 2 hr in the present case does
study shows that the magnetofriction method can also
not directly lead to an unstable state of the system because it
qualitatively emulate the confined partial eruption and the
completely decays about an hour before the onset of the
observed twist evolution (due to the reconnection with
precursor flare.
overlying flux) in a filament. However, cautions must be taken
The double-decker structure of the filament system, i.e., the
in applying the method and in interpreting the modeling result.
trapping of the filaments in two separate magnetic structures, is
For instance, because the method cannot describe any MHD
indicated by the different dynamic behavior of the filaments
waves (fast mode, Alfvén mode, nor slow mode waves), it
during both the precursor activities (intermittent ejections
works as a reliable approximation of the MHD evolution only
disturbing only the lower filament; Section 3.1.1) and the main
when the dynamics are well subsonic and sub-Alfvénic, in
eruption (reconnection between the filaments, producing bright
particular, for quasi-static evolutions. When eruptive dynamics
threads between them; Section 3.2). This is further supported
with strong acceleration of plasma are considered, a careful
by the successful model that employs a double-decker
comparison between the magnetofrictional evolution and the
structure.
observational data or an MHD model is still required for a
To understand how the eruption initiates and why it fails, we
representative number and range of case studies. Our
construct magnetic field models using the flux rope insertion
magnetofrictional modeling yields magnetic reconnection of
method that involves magnetofrictional relaxation. The mag-
erupting flux in qualitative agreement with the observed twist
netofrictional evolution has no intrinsic timescale, as the
release from the erupting flux and the brightening of the main
velocities used in the magnetofrictional scheme are not real
flare ribbons, but it does not yield a long, thin current sheet
fluid velocities. However, the spatial structures, including the
between the erupting and overlying flux systems (Guidoni et al.
magnetic topology, are realistic and even data constrained; thus
2016), similar to previous applications of the method (Su et al.
the resultant magnetic configuration can help understand the
2011; Savcheva et al. 2016; Liu et al. 2018d). This suggests
observations (Cheung & DeRosa 2012; Savcheva et al.
that such modeling yields the large-scale consequences but not
2015, 2016; Guo et al. 2016).
the detailed dynamics of reconnection.
Topological analysis of the potential field shows that the
The main results of our observational and modeling analysis
overlying field of the filament system possesses a QSL of
of the failed eruption can be summarized as follows:
dome-plate geometry but without an embedded magnetic null
point. A high-Q plate separates the dome into two parts. Our 1. The double-decker filament is structured as a flux rope
modeling of the nonpotential coronal field reproduces the atop a flux bundle with negligible twist.
observed failed eruption of the upper filament, with the lower 2. Prior to the eruption, one of the strongest intermittent
filament remaining essentially undisturbed. This includes the ejections from an area of flux cancellation in the filament
accumulation and subsequent release of twist in the upper channel appears to transfer flux into the upper filament,
filament during the eruption. The magnetic configurations after which weakens its stability. Additionally, magnetic
5000, 30,000, and 90,000 magnetofrictional iterations rather reconnection of the ambient field in the space between
closely match the observed filament system before, during, and the two filaments, indicated by brightened threads
after the eruption, respectively. The evolving model also shows immediately following the precursor flare, transfers
a decrease of free magnetic energy, as required for an eruption. further flux to the upper filament and simultaneously
After 5000 iterations, the magnetic field evolves into a reduces the constraint imposed by the overlying field.
configuration with a flux rope atop a flux bundle with Both mechanisms contribute to the observed slow rise

14
The Astrophysical Journal, 923:142 (16pp), 2021 December 20 Chen et al.

and destabilization of the filament, which eventually Awasthi, A. K., Liu, R., Wang, H., Wang, Y., & Shen, C. 2018, ApJ, 857, 124
erupts, once reaching the critical height of the torus Awasthi, A. K., Liu, R., & Wang, Y. 2019, ApJ, 872, 109
instability. On the other hand, the ∼2 hr filament Barczynski, K., Aulanier, G., Masson, S., & Wheatland, M. S. 2019, ApJ,
877, 67
oscillation before the eruption occurs in a stable state of Bateman, G. 1978, MHD Instabilities (Cambridge, MA: MIT Press)
the system is unlikely to trigger the eruption. Berger, M. A., & Field, G. B. 1984, JFM, 147, 133
3. Both observations and magnetic modeling indicate that Berger, M. A., & Prior, C. 2006, JPhA, 39, 8321
the erupted flux is considerably twisted, but the twist does Bobra, M. G., van Ballegooijen, A. A., & DeLuca, E. E. 2008, ApJ, 672, 1209
Chen, J., Liu, R., Liu, K., et al. 2020, ApJ, 890, 158
not reach the threshold of the helical kink instability. The Chen, P. F., Innes, D. E., & Solanki, S. K. 2008, A&A, 484, 487
filament threads trace preexisting twist, which becomes Cheung, M. C. M., & DeRosa, M. L. 2012, ApJ, 757, 147
apparent during the expansion and disintegration of the Démoulin, P., & Aulanier, G. 2010, ApJ, 718, 1388
filament. Additional twist is added to the erupting flux in Demoulin, P., Henoux, J. C., Priest, E. R., & Mandrini, C. H. 1996, A&A,
the course of the rise as a result of tether-cutting 308, 643
DeVore, C. R., & Antiochos, S. K. 2008, ApJ, 680, 740
reconnection of ambient flux under the rising flux. Once Fan, Y., & Gibson, S. E. 2003, ApJL, 589, L105
reaching the overlying dome-shaped QSL, the twist of the Fan, Y., & Gibson, S. E. 2004, ApJ, 609, 1123
upper filament is transferred to the post-flare arcade field Fan, Y., & Gibson, S. E. 2007, ApJ, 668, 1232
whose footpoints are highlighted by the flare ribbons Filippov, B. 2020, MNRAS, 494, 2166
Gary, G. A. 2001, SoPh, 203, 71
surrounding the lower filament. The transfer results from Gibson, S. E., Fletcher, L., Del Zanna, G., et al. 2002, ApJ, 574, 1021
magnetic reconnection with the overlying field. Both the Gilbert, H. R., Alexander, D., & Liu, R. 2007, SoPh, 245, 287
rise of the upper filament and its subsequent untwisting Gilbert, H. R., Holzer, T. E., Burkepile, J. T., & Hundhausen, A. J. 2000, ApJ,
lower the free magnetic energy of the configuration. 537, 503
Guidoni, S. E., DeVore, C. R., Karpen, J. T., & Lynch, B. J. 2016, ApJ, 820, 60
4. The confinement of the eruption is caused by one or both
Guo, Y., Ding, M. D., Schmieder, B., et al. 2010, ApJL, 725, L38
of the following effects. (a) Motion of the erupting flux Guo, Y., Xia, C., & Keppens, R. 2016, ApJ, 828, 83
from a torus-unstable to a torus-stable height range, the Hassanin, A., & Kliem, B. 2016, ApJ, 832, 106
latter of which is located just above the intersection of the Hood, A. W., & Priest, E. R. 1979, SoPh, 64, 303
dome- and plate-shaped QSLs. (b) Erosion of the erupted Hundhausen, A. 1999, in The Many Faces of the Sun: A Summary of the
Results from NASAʼs Solar Maximum Mission, ed. K. T. Strong et al. (New
flux rope by reconnection with overlying flux at the York: Springer), 143
dome-shaped QSL. This is suggested by the draining of Ishiguro, N., & Kusano, K. 2017, ApJ, 843, 101
the filament to partly new footpoint locations, which Janvier, M., Savcheva, A., Pariat, E., et al. 2016, A&A, 591, A141
immediately follows the termination of its rise, and Ji, H., Wang, H., Schmahl, E. J., Moon, Y. J., & Jiang, Y. 2003, ApJL,
595, L135
directly observed in the model. Kliem, B., Lee, J., Liu, R., et al. 2021, ApJ, 909, 91
Kliem, B., & Török, T. 2006, PhRvL, 96, 255002
The authors thank the SDO and the GONG teams for Kliem, B., Török, T., Titov, V. S., et al. 2014, ApJ, 792, 107
providing the valuable data. We thank Jun Chen for kindly Kusano, K., Iju, T., Bamba, Y., & Inoue, S. 2020, Sci, 369, 587
offering the code package for calculating Tw and the Q factor Lemen, J. R., Title, A. M., Akin, D. J., et al. 2012, SoPh, 275, 17
Li, S., Su, Y., Zhou, T., et al. 2017, ApJ, 844, 70
and Drs. Yang Guo and Xin Cheng for valuable discussions. Liu, C., Cao, W., Chae, J., et al. 2018a, ApJ, 869, 21
We also gratefully acknowledge constructive comments by the Liu, L., Cheng, X., Wang, Y., et al. 2018b, ApJL, 867, L5
referee which have stimulated a deeper analysis, especially of Liu, R. 2020, RAA, 20, 165
the reconnection in the model and of the differences in the Liu, R., Chen, J., & Wang, Y. 2018c, SCPMA, 61, 69611
magnetofrictional and MHD descriptions of eruptive processes. Liu, R., Kliem, B., Titov, V. S., et al. 2016, ApJ, 818, 148
Liu, R., Kliem, B., Török, T., et al. 2012, ApJ, 756, 59
This work is supported by the following Chinese foundations: Liu, T., Su, Y., Cheng, X., van Ballegooijen, A., & Ji, H. 2018d, ApJ, 868, 59
the National Natural Science Foundation of China (NSFC; Masson, S., Pariat, E., Aulanier, G., & Schrijver, C. J. 2009, ApJ, 700, 559
41761134088, 11473071, 11790302 (11790300), 11773079, Myers, C. E., Yamada, M., Ji, H., et al. 2015, Natur, 528, 526
41774150, 11925302, and U1731241), the Strategic Priority Pan, H., Liu, R., Gou, T., et al. 2021, ApJ, 909, 32
Parker, E. N. 1974, ApJ, 191, 245
Research Program on Space Science, and Chinese Academy of Pesnell, W. D., Thompson, B. J., & Chamberlin, P. C. 2012, SoPh, 275, 3
Sciences (grant No. XDA15052200 and XDA15320301). B.K. Pevtsov, A. A. 2002, SoPh, 207, 111
acknowledges support by the DFG in the joint Sino-German Plunkett, S. P., Vourlidas, A., Šimberová, S., et al. 2000, SoPh, 194, 371
program NSFC41761134088/DFG KL817.8-1. Priest, E. R. 2014, Magnetohydrodynamics of the Sun (Cambridge: Cambridge
Univ. Press)
Savcheva, A., Pariat, E., McKillop, S., et al. 2015, ApJ, 810, 96
ORCID iDs Savcheva, A., Pariat, E., McKillop, S., et al. 2016, ApJ, 817, 43
Schou, J., Scherrer, P. H., Bush, R. I., et al. 2012, SoPh, 275, 229
Yingna Su https://orcid.org/0000-0001-9647-2149 Su, Y., Liu, R., Li, S., et al. 2018, ApJ, 855, 77
Rui Liu https://orcid.org/0000-0003-4618-4979 Su, Y., Surges, V., van Ballegooijen, A., DeLuca, E., & Golub, L. 2011, ApJ,
Bernhard Kliem https://orcid.org/0000-0002-5740-8803 734, 53
Su, Y., van Ballegooijen, A., McCauley, P., et al. 2015, ApJ, 807, 144
Qingmin Zhang https://orcid.org/0000-0003-4078-2265 Su, Y.-n 2019, ChA&A, 43, 305
Haisheng Ji https://orcid.org/0000-0002-5898-2284 Tang, F. 1986, SoPh, 105, 399
Tie Liu https://orcid.org/0000-0001-7385-4742 Titov, V. S., Hornig, G., & Démoulin, P. 2002, JGRA, 107, 1164
Török, T., & Kliem, B. 2005, ApJL, 630, L97
Török, T., Kliem, B., & Titov, V. S. 2004, A&A, 413, L27
References van Ballegooijen, A. A. 2004, ApJ, 612, 519
van Ballegooijen, A. A., Priest, E. R., & Mackay, D. H. 2000, ApJ, 539, 983
Alexander, D., Liu, R., & Gilbert, H. R. 2006, ApJ, 653, 719 van Tend, W., & Kuperus, M. 1978, SoPh, 59, 115
Amari, T., Canou, A., Aly, J.-J., Delyon, F., & Alauzet, F. 2018, Natur, Wang, D., Liu, R., Wang, Y., et al. 2017, ApJL, 843, L9
554, 211 Wang, H., & Liu, C. 2012, ApJ, 760, 101
Aulanier, G., & Dudík, J. 2019, A&A, 621, A72 Wang, H., Liu, C., Deng, N., et al. 2014, ApJL, 781, L23
Aulanier, G., Török, T., Démoulin, P., & DeLuca, E. E. 2010, ApJ, 708, 314

15
The Astrophysical Journal, 923:142 (16pp), 2021 December 20 Chen et al.

Wang, S., Liu, C., Liu, R., et al. 2012, ApJL, 745, L17 Zheng, R., Yang, S., Rao, C., et al. 2019, ApJ, 875, 71
Yang, S., & Zhang, J. 2018, ApJL, 860, L25 Zhu, C., & Alexander, D. 2014, SoPh, 289, 279
Yang, S., Zhang, J., Song, Q., Bi, Y., & Li, T. 2019, ApJ, 878, 38 Zhu, C., Liu, R., Alexander, D., Sun, X., & McAteer, R. T. J. 2015, ApJ,
Yang, W. H., Sturrock, P. A., & Antiochos, S. K. 1986, ApJ, 309, 383 813, 60

16