Supra-arcade Downflows: Dynamics & Diagnostics

• Supra-arcade downflows (SADs) are dark, sunward-moving, density-depleted voids in the hot supr-arcade fan that signal active magnetic reconnection processes.
• Observations reveal SADs with distinct tadpole- or finger-like shapes, speeds of 45–500 km/s, deceleration, and variations in thermal properties quantified via DEM analyses.
• Competing models interpret SADs as retracting flux tubes, reconnection outflows, or RT instabilities, with recent giant filament eruptions showing similar phenomena across different scales.

Supra-arcade downflows (SADs) are dark, sunward-moving, low-emission structures observed above post-eruption flare arcades within the hot supra-arcade fan or plasma sheet. In EUV and soft X-rays they commonly appear as tadpole-like or finger-like voids descending through bright, high-temperature plasma, and they are closely associated with magnetic reconnection in the current sheet above the arcade. Although SADs were historically studied mainly in active-region flares, the December 24, 2023 giant quiescent solar filament eruption demonstrated that they also occur in much larger, weaker-radiation systems: the supra-arcade fan in that event was about three times as wide as comparable active-region fans, the largest SAD widths were at least three times larger than the maximum widths reported in active-region flares, the velocities remained comparable to typical active-region values, and the event showed lower temperature, lower emission measure, less overall radiation, and no evident GOES signature (Xie et al., 12 Sep 2025).

1. Observational definition and phenomenology

SADs are observed as dark, plasma-depleted voids above flare loops, descending toward the top of the post-flare arcade through the hot supra-arcade fan. Their appearance as intensity depressions follows directly from optically thin emission, for which the observed signal scales as

$$I_\lambda = \int R_\lambda(T)\,\mathrm{DEM}(T)\,\mathrm{d}T$$

or, equivalently in imaging terms, as

$I \propto \int n_e^2 G(T)\,dl.$

A reduction in emission measure, a temperature shift relative to the instrumental response, or both can therefore produce a dark lane against a bright fan (Chen et al., 2017).

Morphologically, SADs are often described as thin, collimated, sinuous “tadpoles,” with a darker head and a trailing tail, although finger-like and teardrop-shaped forms are also common. Supra-arcade fans themselves are bright, irregular regions of hot plasma above flare arcades, best seen in hot AIA channels such as 131 Å, and can extend up to roughly 150 Mm above the arcade and persist for hours. Within such fans, SADs may show head splitting, pronged tips, narrowing during descent, or association with upward-growing bright spikes (Innes et al., 2014).

Published observational ranges are broad but internally consistent. Typical reported SAD properties include emission deficits consistent with densities 5–43 times lower than the surrounding fan or arcade plasma, cross-sectional areas of a few to $70 \times 10^6\ \mathrm{km}^2$, effective radii of 1–4 Mm, sunward speeds of 45–500 km s$^{-1}$, decelerations around 1.5 km s$^{-2}$, lifetimes of a few minutes, origin heights of 100–200 Mm above the limb, and penetration depths of 20–50 Mm into the fan region. Multi-instrument statistical work further found cross-sectional areas of about 2–90 Mm$^2$, with at least 75% smaller than 40 Mm$^2$, reinforcing that SADs are typically small relative to the scale of the flare arcade [(Cassak et al., 2013); (Savage et al., 2012)].

Occurrence is most frequent in the decay phase of long-duration eruptive flares, but SADs are not confined to that phase. They may begin earlier, correlate with hard X-ray bursts, and appear together with reconnection inflows and outflows near current sheets. The December 24, 2023 quiescent-filament event established that the phenomenon is not restricted to compact active-region flares, but extends to giant filament eruptions with much broader supra-arcade systems [(Cassak et al., 2013); (Xie et al., 12 Sep 2025)].

2. Kinematics and spatiotemporal behavior

Observed SAD motions are systematically sub-Alfvénic. In one single-flare statistical study of 81 SADs, projected appearance heights were about 50–200 Mm, disappearance heights about 20–130 Mm, initial projected velocities about 100–400 km s$^{-1}$, and initial accelerations about $-2$ to $0$ km s$I \propto \int n_e^2 G(T)\,dl.$0. The trajectories were not uniform: 31 were fit by a single constant deceleration, 28 by two-stage deceleration, and 22 exhibited at least one phase of positive acceleration during descent (Tan et al., 2022).

Detailed thermodynamic tracking of 20 SADs in three M-class flares yielded initial speeds of order 100 km s$I \propto \int n_e^2 G(T)\,dl.$1, quasi-monotonic deceleration to approximately zero at the loop top, and a mean deceleration of about 0.15 km s$I \propto \int n_e^2 G(T)\,dl.$2. Those values fall within the lower half of the broader SAD speed range from earlier imaging studies and are consistent with the long-standing inference that SADs experience substantial drag or pressure braking as they descend into denser plasma above the arcade (Li et al., 2021).

Modern spectroscopy confirms that the downflows are genuine moving structures rather than merely stationary emissivity deficits. Hinode/EIS and AIA measurements for four prominent SADs in the 2022 April 2 flare gave line-of-sight velocities of $I \propto \int n_e^2 G(T)\,dl.$3, $I \propto \int n_e^2 G(T)\,dl.$4, $I \propto \int n_e^2 G(T)\,dl.$5, and $I \propto \int n_e^2 G(T)\,dl.$6 km s$I \propto \int n_e^2 G(T)\,dl.$7, plane-of-sky velocities of $I \propto \int n_e^2 G(T)\,dl.$8, $I \propto \int n_e^2 G(T)\,dl.$9, $70 \times 10^6\ \mathrm{km}^2$0, and $70 \times 10^6\ \mathrm{km}^2$1 km s$70 \times 10^6\ \mathrm{km}^2$2, and 3D speeds of $70 \times 10^6\ \mathrm{km}^2$3, $70 \times 10^6\ \mathrm{km}^2$4, $70 \times 10^6\ \mathrm{km}^2$5, and $70 \times 10^6\ \mathrm{km}^2$6 km s$70 \times 10^6\ \mathrm{km}^2$7. The same study reported a north–south asymmetry in Doppler signatures, interpreted as divergence above the flare loop arcade and possibly related to a high-altitude termination shock (French et al., 28 May 2025).

Temporal organization is also nontrivial. Waiting-time analyses over seven eruptive flares found that most SAD waiting-time distributions are best fit by a power law with slope ranging from 1.7 to 2.4, although log-normal fits can be nearly as good. These distributions rule out linear random or quasi-periodic generation and instead indicate clustered, heavy-tailed timing behavior in the reconnection outflow region (Yang et al., 22 Jul 2025).

3. Thermal structure, emission measure, and plasma diagnostics

Differential emission measure (DEM) analysis is the standard method for quantifying SAD thermodynamics. With

$70 \times 10^6\ \mathrm{km}^2$8

one can separate the hot supra-arcade component from the cooler line-of-sight background and compare plasma inside the SAD, at its front, and in the surrounding fan (Li et al., 2021).

A robust observational result is that SADs are primarily density-depleted structures. In two limb flares analyzed with AIA DEM inversion, the hot DEM above 4 MK decreased by 1–3 orders of magnitude during SAD passage relative to the surrounding haze and to the same region before or after passage, while the DEM-weighted temperature remained broadly in the 12–15 MK range within uncertainties. No significant DEM component above 20 MK was detected in those AIA data, and the depressed hot DEM typically recovered within a few minutes, indicating that SADs are discrete, transient voids rather than long-lived evacuated channels at a fixed location (Chen et al., 2017).

Time-dependent DEM studies refine that picture. In 20 tracked SADs, both the front and the main body tended to heat during descent, the front was typically hotter than the body, and the high-temperature DEM component strengthened as the structures moved downward. The DEM-weighted temperature increased by more than 30% in most cases and approached about 10 MK near the loop top. A separate single-flare study of 54 SADs found that both front and body temperatures increased during descent and that the body generally remained underdense relative to the front and the fan (Li et al., 2021, Tan et al., 2022).

Spectroscopy shows that darkness need not imply a much lower temperature than the fan. In the 2022 April 2 event, temperatures derived from the Fe XXIV 255.11 Å / Fe XXIII 263.41 Å ratio were $70 \times 10^6\ \mathrm{km}^2$9, $^{-1}$0, $^{-1}$1, and $^{-1}$2 MK for four SADs, compared with $^{-1}$3 MK in a nearby non-SAD fan region. The same analysis found enhanced non-thermal velocities of $^{-1}$4, $^{-1}$5, $^{-1}$6, and $^{-1}$7 km s$^{-1}$8, compared with $^{-1}$9 km s$^{-2}$0 in the surrounding fan, implying unresolved or turbulent motions associated with the downflows (French et al., 28 May 2025).

Thermal behavior can vary with flare phase and geometry. In six on-disk SAD episodes on 2013 April 11, cases occurring close to flare maximum contained an enhanced 5–7 MK component within the void, whereas later cases exhibited a depression in hot plasma at 7–12 MK relative to adjacent supra-arcade rays. That progression was interpreted as an evolution from hotter early downflows to increasingly density-depleted later ones (Awasthi et al., 2022).

4. Physical interpretations and competing models

The origin of SADs has been debated since their discovery, and the literature contains several physically distinct interpretations. One early framework treated SADs as the cross-sections of newly reconnected flux tubes retracting sunward from reconnection sites high in the corona, with supra-arcade downflowing loops (SADLs) representing the same objects seen from a different line of sight. Multi-wavelength statistical work supported this geometrical link by showing comparable kinematics and by estimating fluxes of order $^{-2}$1 Mx per tube and shrinkage energies of order $^{-2}$2 erg (Savage et al., 2012).

A major reinterpretation followed from high-cadence AIA observations of the 2011 October 22 flare. In that event, thin bright shrinking loops were seen at the leading edge of much larger descending voids, with a highlighted case showing a void head about 20 pixels wide, or about 9 Mm, while each leading loop was only about 2–3 pixels wide, or $^{-2}$3–$^{-2}$4 Mm. DEM maps showed the emission measure within SADs depressed by a factor of about 4 in the 10–13 MK range, with no compensating enhancement at lower temperatures. These observations motivated the proposal that many SADs are not tube cross-sections but wakes or trailing density depletions behind much thinner retracting loops (Savage et al., 2011).

A different explanation emphasizes continuous reconnection outflows. In that model, spatially localized but temporally continuous reconnection in a stratified corona drives sunward jets of lower-density plasma into the denser supra-arcade region, carving collimated depletion channels. Density stratification is essential, continuous jetting prevents rapid refill from behind, and three-dimensional localization with a guide field keeps the channels thin. A simple flux-conservation scaling gives

$^{-2}$5

so for $^{-2}$6 and $^{-2}$7–60 Mm, one obtains $^{-2}$8–4 Mm, consistent with observed SAD radii (Cassak et al., 2013).

A fourth major class of models invokes secondary Rayleigh–Taylor-type or interchange-like instabilities in the reconnection exhaust. In this picture, a broad low-density reconnection outflow impinges on denser plasma accumulated above the arcade, creating a decelerating interface that is unstable to finger formation. The classical growth rate may be written

$^{-2}$9

with Atwood number $^2$0, and magnetic tension modifies it to

$^2$1

Three-dimensional MHD simulations showed low-density descending fingers intertwined with bright spikes, head splitting, and spike-apex formation in a manner resembling AIA observations [(Innes et al., 2014); (Guo et al., 2014)].

The 2023 quiescent-filament eruption strongly supports the relevance of system-scale RT-like scaling. Its largest SADs were at least three times wider than the largest active-region SADs, but the velocities remained similar to active-region values. The event was interpreted as consistent with previous predictions that larger systems permit larger-scale instabilities while leaving the velocity development largely independent of system size (Xie et al., 12 Sep 2025).

5. Transport processes, heating, and interaction with the arcade

SADs do not simply traverse a passive background. DEM and flow analyses of the 2011 October 22 supra-arcade plasma sheet found that adiabatic compression ahead of SADs produces heating of order 0.1–0.2 MK s$^2$2, comparable to estimated conductive cooling rates, while viscous heating is about an order of magnitude smaller. Ten of the 13 tracked SADs displayed clear signatures in both adiabatic and viscous terms, indicating that SAD motion actively modifies local thermodynamics in the fan (Reeves et al., 2017).

Independent thermodynamic studies reached a similar conclusion from a different diagnostic. For both the front and the body, the relation between pressure and temperature followed the adiabatic expectation

$^2$3

with $^2$4. Across 20 SADs, the fitted $^2$5–$^2$6 slopes had mean values of about 0.44 for the front and 0.41 for the body, close to the theoretical $^2$7, implying that adiabatic compression is the dominant heating mechanism during descent (Li et al., 2021).

Thermal conduction poses a serious theoretical constraint on SAD survival. Two-dimensional MHD simulations including anisotropic conduction found that for a representative SAD diameter of about 1.2 Mm and a fan temperature of about 7 MK, the parallel conductive timescale is about 0.5 s, far shorter than observed SAD lifetimes. The simulations therefore concluded that long-lived SADs require magnetic insulation: field-line wrapping or magnetic islands must suppress parallel heat flux and preserve the emission-measure deficit (Zurbriggen et al., 2017).

On-disk observations further showed that SAD interaction with the arcade can excite large-scale wave responses. In the 2013 April 11 flare, six SAD episodes produced transverse oscillations in supra-arcade rays with periods of about 120–160 s, footpoint brightenings in 1700 Å delayed by 22–46 s relative to SAD arrival at the arcade top, and EUV intensity perturbations propagating at approximately 400 km s$^2$8. DEM analysis indicated that the interaction did not produce significant hot looptop heating; rather, it mainly launched MHD disturbances directed toward the lower atmosphere (Awasthi et al., 2022).

6. Statistical properties, scaling relations, and broader significance

SAD populations exhibit nontrivial statistical structure. A Yohkoh/SXT sample of 120 SADs found that cross-sectional areas are consistent with a log-normal distribution, while magnetic fluxes are consistent with both log-normal and exponential distributions; neither areas nor fluxes are compatible with power-law or normal distributions. This was interpreted as evidence against a purely scale-free fractal creation process and as support for multiplicative growth scenarios in patchy reconnection (McKenzie et al., 2011).

Width evolution within a single flare also shows systematic behavior. For 81 SADs observed on 2013 May 22, initial widths spanned about 4.6–24.9 arcsec, and most structures showed a roughly monotonous width decrease during descent, although about half displayed more complex evolution with intermittent broadening. Three-dimensional reconstruction with SDO and STEREO-A indicated that geometric and environmental effects matter: mean maximum width was about 16.2 arcsec in the north and about 11.9 arcsec in the south, with corresponding loop angles to the line of sight of about $^2$9 and $^2$0. Smaller-initial-width SADs preferentially appeared at lower heights, where more frequent collisions were associated with intermittent acceleration, width increases, and curved trajectories (Tan et al., 2023).

Temporal statistics convey comparable information. Waiting-time studies over seven eruptive flares found heavy-tailed distributions, with power-law slopes around 1.7–2.4 and frequent near-degeneracy with log-normal fits. These results are inconsistent with stationary Poisson or quasi-periodic generation and instead point to intermittent, nonlinear dynamics in the reconnection outflow region, plausibly involving coupled SOC-like triggering, turbulence, and plasmoid-mediated variability (Yang et al., 22 Jul 2025).

The 2023 quiescent-filament eruption extends these statistical and physical inferences to a different eruption class. The fan was about three times as wide as similar structures above active regions, the largest SAD widths were at least three times larger than the largest active-region SADs, yet the velocities remained comparable to typical active-region values. At the same time, the event exhibited lower temperature, lower emission measure, less overall radiation, and no evident GOES signature, while the SADs themselves remained cooler than the surrounding fan plasma. A plausible implication is that SAD formation depends more fundamentally on reconnection-exhaust structure and instability development than on the compactness or radiative brightness of the eruption, so that quiescent filament eruptions and active-region flares can generate homologous SAD phenomenology across very different global scales (Xie et al., 12 Sep 2025).