THERMAL STRUCTURE OF CURRENT SHEETS AND SUPRA-ARCADE DOWNFLOWS IN THE SOLAR CORONA

Will J. Hanneman; Katharine K. Reeves

doi:10.1088/0004-637X/786/2/95

1. INTRODUCTION

On 1999 January 20 at 1900 UT, a flare and coronal mass ejection were observed on the northeast limb of the Sun by the Soft X-ray Telescope (SXT) on the Yohkoh satellite (McKenzie & Hudson 1999). Roughly 2 hr after the flare began, dark features descended upon the arcade, perturbing the fan-shaped structures that appeared above the post-flare loops. These dark features have been identified in several other flares since then (McKenzie 2000; Innes et al. 2003; Asai et al. 2004; Sheeley et al. 2004; Verwichte et al. 2005; Khan et al. 2007; Reeves et al. 2008; McKenzie & Savage 2009; Savage & McKenzie 2011; Warren et al. 2011; Savage et al. 2012; Liu 2013; Liu et al. 2013), and have come to acquire the name supra-arcade downflows (SADs; although some papers refer to them as dark lanes or dark voids). Asai et al. (2004) find that the SADs observed during the impulsive phase of a flare are correlated with hard X-ray bursts, which are caused by particles being accelerated by the reconnection process. Savage & McKenzie (2011) find that the SADs generally have faster velocities at the start of the flare, and decelerate toward the top of the flare arcade, as would be expected for a reconnecting flux rope. It is currently thought, therefore, that SADs are a product of the reconnection process.

While the kinematics of SADs have been well studied, less is known about the plasma characteristics within the SADs. Spectroscopic measurements have shown that SADs are sunward-flowing regions of reduced density (Innes et al. 2003), a claim that has recently been confirmed using differential emission measures (DEMs) calculated from imaging data (Savage et al. 2012). While there is general agreement as to the very low density of these voids, the temperatures have not been well studied. McKenzie & Hudson (1999) used filter ratios of SXT data to determine temperatures, and found that the voids in their event were slightly hotter than the surrounding fan-shaped supra-arcade plasma. The signal in the voids is very weak in the X-rays, however, so these measurements are not conclusive.

Several physical interpretations of the causes of SADs have been presented. Linton & Longcope (2006) and Guidoni & Longcope (2011) provide a patchy reconnection scenario that could be responsible for the SADs. A recent observational interpretation is that they are the wakes of thin shrinking loops (Savage et al. 2012). Modeling efforts that mirror this interpretation have recently been presented (Scott et al. 2013). In this model, a flux tube retracts, constricting the surrounding field and causing a peristaltic pumping action that shocks the plasma in front of the retracting tube and leaves a void in its wake. Maglione et al. (2011) and Cécere et al. (2012) present a different interpretation than Savage et al. (2012) and Scott et al. (2013). In their model, SADs are created by a pressure pulse, which is caused by the reconnection of field lines. This pressure pulse is coupled with a perpendicular magnetic shock and causes a cavity to form, resulting in SADs. This model predicts that the temperatures inside the SADs should be significantly hotter than the surrounding plasma. Another recent model posits that the SADs are channels caused by sunward flowing reconnection jets (Cassak et al. 2013). This model predicts that the SADs are of a lower temperature than the surrounding plasma (P. Cassak 2013, private communication). Because these models predict different temperature profiles for the plasma in the downflow region when compared to the supra-arcade fan plasma, it is clear that measurements are needed in order to clarify the mechanism responsible for the SADs.

In this paper, we measure the plasma temperature of the SAD regions and the surrounding plasma sheet using data from the Atmospheric Imaging Assembly (AIA) on the Solar Dynamics Observatory (SDO) and the X-Ray Telescope (XRT) on Hinode. Our instrumentation and methods are presented in Section 2. Our results and analysis are presented in Section 3, and discussion and conclusions are presented in Section 4.

2. INSTRUMENTATION AND METHOD

2.1. AIA

In 2010 February, as part of NASA's "Living With a Star" program, the SDO was launched into orbit on a five-year mission to help our understanding of the Sun's influence on Earth (Pesnell et al. 2012). The AIA is one of three instruments on board the spacecraft (Lemen et al. 2011). Taking mostly narrowband images in extreme ultraviolet (EUV) wavelengths, AIA has a cadence of 12 s and a spatial resolution of 0 farcs 6 pixel⁻¹. The wavebands used in this paper are the six that measure emission predominately from iron lines—94 Å, 131 Å, 171 Å, 193 Å, 211 Å, and 335 Å. The temperature sensitivities range from 4 × 10⁵ K (Fe viii) to 2 × 20⁷ K (Fe xxiii), although sensitivities at T > 10 MK are relatively low (O'Dwyer et al. 2010; Del Zanna et al. 2011).

AIA response functions are calculated using the function aia_get_response, which is available in the SolarSoftWare (SSW) IDL software database (Freeland & Handy 1998). We used version 4 of the AIA calibration, the most recent at the time of this writing (see Boerner et al. 2014, for details). This version incorporates lines from the latest version of the CHIANTI database (Version 7.1; Dere et al. 1997; Landi et al. 2013). We also used the options that account for missing lines in CHIANTI (/chiantifix) and normalize the response functions to output from the Extreme Ultraviolet Variability Experiment (EVE) on SDO (/evenorm).

The AIA instrument consists of four telescopes with two channels each, and there can be blending between the two channels on some of the telescopes. Because of the properties of the mirrors, the most significant effect is in the 335 Å channel (Boerner et al. 2012). The default temperature response functions of this channel take into account the blending, which should result in some radiation from the 131 Å passing through the 335 Å channel. Observational analysis in recent papers suggests this contribution is overestimated in the standard response functions (McCauley et al. 2013), and better results are obtained by ignoring the contribution to the 335 Å channel from the 131 Å channel. Therefore, we used an option in the standard AIA response function software (/noblend) that acts to alter the temperature response functions to eliminate the blending effect. This option was used throughout this paper and is justified in Section 3.2.

2.2. XRT

Hinode is a joint NASA/Japan Aerospace Exploration Agency Solar mission launched in 2006 September (Kosugi et al. 2007). One of the three instruments on board is the X-Ray Telescope (XRT; Golub et al. 2007). XRT has broadband filters that detect temperatures above 2 MK (Narukage et al. 2011) with a spatial resolution of 1 farcs 0286 per pixel. Although XRT is capable of full-Sun imaging, it typically takes images with a small field of view, usually 384''×384'', in response to solar flares. XRT has an automatic flare detection system for capturing flares (Kano et al. 2008), but it only captures events within one solar radius of the pre-flare pointing. Thus XRT data is not always available for solar events, including for two of the four flares studied here.

The XRT filters used in this paper are Ti-poly, Be-thin, Al-med, Be-med, Al-thick, and Be-thick. We use XRT temperature response functions that are calculated using the same CHIANTI Version 7.1 spectra as the AIA temperature response functions. The instrument characteristics used in calculating these response functions use the most recent calibrations for the filter thicknesses, as described in Narukage et al. (2014).

2.3. Data Preparation and Combination

Four recent flare events with large SADs have been analyzed here. Plots of the 1–8 Å X-ray flux from the Geostationary Operational Environmental Satellite (GOES) are shown for each flare in Figure 1. The flares took place on 2011 October 22 at 11:10 UT (GOES Class M1.3), 2012 January 14 at 13:18 UT (GOES Class M1.4), 2012 January 16 at 04:44 UT (GOES Class C6.5), and 2012 January 27 at 18:37 UT (GOES Class X1.7). Three of the four flares are long duration events. The fourth flare, which occurred on 2012 January 14, is an impulsive event that occurred in the same region as a previously occurring long duration event.

**Figure 1.** Plots of the *GOES* 1–8 Å flux for the flares in this study. Vertical bars indicate the times at which DEMs are calculated. Panel (a) *GOES* Class M1.3 flare on 2011 October 22 at 11:10 UT. Panel (b) *GOES* Class M1.4 flare on 2012 January 14 at 13:18 UT. Note that the flare of interest is the impulsive peak, rather than the more long duration structure. Panel (c) *GOES* Class C6.5 flare on 2012 January 16 at 04:44 UT. Panel (d) *GOES* Class X1.7 flare on 2012 January 27 at 18:37 UT.
Download figure:
Standard image High-resolution image

For each event, we choose a time where the SADs are clearly visible in an AIA 131 Å image. Further criteria are that no visible saturation or diffraction patterns from the flare are visible in the 131 Å image or in the other five EUV wavelengths that are nearest in time to the 131 Å image. The six images are then processed by cutting out the part of the full Sun image that contains the flare. A SSW routine called aia_prep is used to normalize the instrument plate scale across the four AIA telescopes, de-rotate the images, and align all the cut-outs to the 131 Å cut-out. The intensity in each filter is also normalized to its exposure time. AIA data are deconvolved with the point-spread function in each filter using the SSW routine aia_deconvolve_richardsonlucy.

Data from XRT is available for the 2012 January 16 and 2012 January 27 flares. A SSW routine called xrt_prep is used on the XRT data to perform standard dark subtraction and to remove the CCD bias and telescope vignetting (see Kobelski et al. 2014, for details of the calibrations performed by xrt_prep). Intensities in XRT filters are also normalized to the exposure time, and the XRT images are de-rotated to account for a slight rotation of the XRT instrument with respect to solar north. The XRT data are also deconvolved with the XRT point-spread function using the same Richardson-Lucy algorithm that is used for the AIA data, but with fewer iterations (we use niter = 7 for XRT, whereas aia_deconvolve_richardsonlucy uses niter = 25).

AIA and XRT data arrays are combined by re-gridding the AIA data by a factor of 7/12 to match the plate scale of XRT and then overlaying the pixels in AIA wavebands with those in XRT. Alignment of AIA and XRT is done by eye and is accurate to about one XRT pixel.

2.4. Differential Emission Measures (DEMs)

In order to get temperature information about the SADs and surrounding fan plasma, we use intensities from multiple filters from the imaging telescopes to generate DEMs. The times at which we calculate the DEMs are shown as vertical bars in Figure 1. We calculate the DEMs in IDL using xrt_dem_iterative2, written by Mark Weber (Weber et al. 2004; Golub et al. 2004; Cheng et al. 2012), which is available in the SSW database. The program employs an iterative process for finding the best-fit DEMs. It starts with a DEM with equal quantities of plasma at all temperatures, and then determines how well observations produced by this DEM fit the measured intensities. A series of spline knots in the DEM are adjusted until the DEM produces the best match to the measured intensities, as determined by minimizing the χ² value of the measured intensities and those calculated with the DEM. This algorithm has been well tested (see the Appendix of Cheng et al. 2012), and it has also been shown by Schmelz et al. (2009a, 2009b) to provide solutions that are in good agreement with other DEM reconstruction methods, such as the Markov Chain Monte Carlo (MCMC) method developed by Kashyap & Drake (1998).

In order to calculate the DEMs, intensities in each filter are needed. To reduce the possibility of anomalous pixels leading to erroneous results, areas of the images, usually square boxes of 49 pixels, are averaged. All the boxes were positioned by eye using the AIA 131 Å image of the flare. Before averaging, each box was checked to make sure there were no anomalous pixels within it. Major sources of anomalous pixels could include cosmic rays or bad CCD pixels. We determine the error in the final average intensity of each box for each wavelength by taking the standard deviation of the pixels that are averaged. Within the constraints of these errors, the average intensities are used in conjunction with AIA and XRT temperature response functions to calculate the DEMs in each part of the flare.

Because we are interested in the temperatures in the SADs relative to the supra-arcade fan plasma, the process was repeated for other parts of the flare as a basis for comparison. Four specific regions (boxes) for each flare were considered: the downflows themselves, the parts of the arcade adjacent to each downflow, areas of background emission, and the brightest part of the flare in 131 Å. Every DEM from an additional downflow was accompanied by a DEM from an adjacent part of the supra-arcade fan (at the same distance from the Sun's surface) so that the downflow DEM could be put into context.

3. RESULTS AND ANALYSIS

3.1. 2011 October 22 Flare

On 2011 October 22 at 10:00 UT, a flare erupted on the northwest limb of the Sun. It reached a magnitude of M1.3 at its peak and lasted for approximately 6 hr. A large number of SADs occurred between 11:30 UT and 14:30 UT, and Figure 2 shows four progressive AIA 131 Å images within the space of 11 minutes for this flare. Unfortunately, no XRT data is available for this flare.

3.1.1. SAD Motion and Characteristics

Figure 2 shows the progression of several SADs as a function of time in the AIA 131 Å channel. The color table has been reversed in these images so that a higher intensity translates to a darker color. These images demonstrate the movement of the downflows toward the surface of the Sun and their subsequent fading as they reach the top of the arcade. At 12:01:11 UT the first of the downflows can be seen emerging at the top of the flare. By 12:05:11 UT all three downflows are visible and still hold their rounded shape. At 12:07:11 UT they each begin thinning out and start to split in half, and by 12:12:11 UT they have almost completely dissipated into the arcade. This flare is the only one in this paper to exhibit round downflows—the SADs in the other flares that we study are much more elongated in comparison.

This event has been analyzed previously by Savage et al. (2012), who find that the speeds of these downflowing features are between 50 and 500 km s⁻¹. Turbulent motions are also seen in the fan plasma, and these motions have been studied in detail by McKenzie (2013), who find evidence of vorticities that vary in time, stagnation points, and numerous velocity shears.

3.1.2. Indications of SAD Temperatures

Figure 3 shows the October 22 flare in four different AIA wavebands: 171 Å, 211 Å, 94 Å, and 131 Å. These wavebands are sensitive to plasmas at 0.8 MK, 2 MK, 6 MK, and 11 MK, respectively, although the 94 Å and 131 Å channels also have some sensitivity at temperatures of ∼1 MK (see Figure 11 in Boerner et al. 2012). All of the images in Figure 3 are taken within 10 s of each other. The supra-arcade plasma in which the downflows are embedded is somewhat visible in 94 Å, very clearly visible in 131 Å, and not at all visible in the other two cooler wavelengths. These observations indicate that the plasma sheet containing the downflows is at a high temperature since there are no traces of fan plasma evident in the 171 Å or 211 Å (see also Reeves & Golub 2011).

The lack of emission in the 171 Å or 211 Å in the area of the SADs is consistent with the idea that the SADs are due to a density depletion rather than a temperature variation. Given the double peak in the 131 Å response function, dark regions in the 131 Å images could be caused by plasma with a similar density as the fan plasma, but with a temperature at the trough between the two peaks. However, no bright structures corresponding to the SADs are seen in the cooler wavelengths. This observation indicates that the lack of emission in the 131 Å and 94 Å images is due to a lack of density in the SADs, rather than a temperature variation over a plasma structure of consistent density.

In order to gain more quantitative insight into the temperatures and emission measures in the fan plasma and SADs in this flare, we calculate DEMs using the six EUV filters on AIA. We choose several regions to perform DEM analysis on, which are plotted as boxes in Figure 4. We choose regions in each of the three big prominent downflows visible at 12:07:11 UT in the 131 Å image, along with corresponding regions in the fan plasma at approximately the same altitude above the Sun's surface. We also choose a background region outside the fan plasma, and a region at the top of the flare arcade, for comparison.

**Figure 4.** AIA 131 Å image of the 2011 October 22 flare showing the locations of the regions used in the DEM calculations. Labels are directly underneath their respective boxes. The color table has been reversed.
Download figure:
Standard image High-resolution image

The DEMs are plotted in Figure 5. The first thing to note about the DEMs is that the arcade part of the flare (labeled ARC) has a relatively high emission measure and a peak temperature of log(T) = 6.9. The background (labeled BG) has a relatively low peak temperature (log(T) = 6.3) compared to the rest of the flare. These results are similar for all other flares studied here. The DEMs for the three SADs and the three regions of fan plasma show that the peak temperatures of all three downflows are lower than those of the corresponding regions of the fan plasma. The DEM for FAN1 peaks at log(T) = 7.0, whereas the DEM for SAD1 peaks at log(T) = 6.9. The DEM for FAN2 also has a peak at log(t) = 7.0. The corresponding SAD2 DEM peaks at log(T) = 6.3, although the DEM falls off more slowly at high temperatures than low temperatures, indicating the presence of some plasma in the log(T) = 6.8–7.0 range. The DEM for FAN3 has a broad shoulder that peaks at log(T) = 7.0, while the DEM for SAD3 is very similar to that of the background plasma, although it is somewhat broader. The DEMs in the SAD regions also show a lower emission measure than their supra-arcade counterparts in the temperature range log(T) = 6.8–7.5. For example, the emission measure at log(T) = 7.0 given by the DEM for FAN1 is about three times as much as the emission measure given by the DEM for SAD1 at the same temperature.

We also compare the observed intensities (F_obs) to the intensities predicted by the DEMs (F_pred) in Figure 5. We present the ratio of these intensities on a log scale so that underpredictions are weighted the same as overpredictions. We also show error bars on this ratio to indicate the range of possible observational values given the uncertainties in the data. The farther away from the solar surface, the less signal in all wavebands, which translates to a greater uncertainty in the measured intensities. Thus SAD3 and FAN3 have the biggest error bars. For this flare, the intensities predicted by the SAD and fan DEMs agree within the errors with observations taken for the most part, and in most cases agree within a factor of 1.5. The DEMs for the fan plasma tend to underpredict the observations in the 94, 131, and 211 Å wavebands, and overpredict it in the 335 Å waveband. There are not any clear trends in the predicted intensities for the SAD plasma.

Monte Carlo simulations are a method of determining how well constrained the DEMs are, given the uncertainties in the observations. Each Monte Carlo simulation is a DEM solution using intensity values that consist of the measured intensities in each filter plus a random value picked from a normal distribution that is centered at zero and has a standard deviation equal to the error of the measured intensity in that filter. The original DEM is the solution that best fits the measured intensity values. For each DEM, 100 Monte Carlo simulations are carried out to give an overall picture of the uncertainties in the DEM. Figure 6 shows the spread of the Monte Carlo simulations for the FAN1 and SAD1 DEMs. Dark gray boxes in each temperature bin encompass 95% of the Monte Carlo solutions, gray boxes encompass 80% of the solutions, and light gray boxes encompass 50% of the solutions. The plot of the SAD1 Monte Carlo runs shows that the SAD1 DEM is well defined at its peak (where log(T) = 6.4), but becomes more uncertain at higher and lower temperatures. At high temperatures, the dark gray boxes that outline 95% of the solutions extend to low DEM values because there are nine Monte Carlo solutions that peak at log(T) = 6.4, and then decrease steadily, without exhibiting the second peak at higher temperatures. These solutions tend to be generated from intensities that are at the extreme low end of the statistical spread of intensities used by the Monte Carlo DEMs in at least two of the AIA filters.

The Monte Carlo simulations have a small spread throughout the temperature range for the FAN1 DEM shown in Figure 6, indicating that the DEM of FAN1 has a relatively small error at all temperatures. The more tightly clustered DEMs in the FAN1 Monte Carlo simulations, as compared to the SAD1 Monte Carlo simulations, is probably the result of a much higher signal in the hot filters and is characteristic of the spread in the Monte Carlo runs for all the DEMs for this flare. The DEMs throughout this paper are almost always very well defined at the peak emission measure, which is an important value because it represents the temperature of the bulk of the plasma contained in the box.

3.2. 2012 January 14 Flare

On 2012 January 14 at 11:00 UT, a long duration magnitude C4.1 solar flare occurred on the northeast limb of the Sun. Then at 13:14 a longer, narrower eruption suddenly emerged very slightly south of the original flare structure, reaching a magnitude of M1.4. Almost immediately a long, forked downflow appeared in this second flare structure and lasted about an hour, by which time both structures had significantly faded. The downflow in this flare did not exhibit the smooth downward motion of those in the 2011 October 22 flare, but had a more wavy side-to-side motion, as pressure from the flare arcade seemed to stop it from traveling sunward. An AIA 131 Å image is shown in Figure 7. The structure with the SADs is in the southern part of the image, and the bright structure in the northern part of the image is from the preceding C4.1 flare.

We calculate DEMs for the boxes shown in Figure 7. Figure 8 shows the DEMs for the SAD (SAD1), corresponding fan plasma (FAN1), the flare arcade from the C4.1 flare (ARC2), and the background box (BG). The ratio of predicted intensities to observed intensities is shown in the bottom panel of Figure 8. The predicted intensities from the DEMs for SAD1 and FAN1 agree quite well with actual observations. The peak emission measure of the DEM for the downflow is at a lower temperature than that of the supra-arcade plasma adjacent to it. The emission measure at log(T) ≳ 7.3 (∼20 MK) is unlikely to be accurate given that the AIA filters are not sensitive to plasma at these temperatures (Boerner et al. 2012). Thus the indication of some extremely hot plasma in the SAD is probably spurious. Evidence of the uncertainty in the DEMs at high temperatures can be seen in the Monte Carlo simulations for SAD1, as shown in Figure 9. This plot shows that the temperature of the peak emission measure is very well defined, but once temperatures exceed 8 MK, the spread in the possible DEM solutions becomes very large.

**Figure 9.** DEMs resulting from Monte Carlo simulations for SAD1 for the 2012 January 14 flare. The DEM calculated from the observed intensities is shown as a black line. The dark gray boxes encompass 95% of the Monte Carlo solutions in each temperature bin, the gray boxes encompass 80% of the solutions, and the light gray boxes encompass 50% of the solutions.
Download figure:
Standard image High-resolution image

In calculating the DEMs throughout this paper, we have removed the contribution of the 131 Å channel to the 335 Å channel in the AIA response functions as mentioned in Section 2.1. For this flare, we compare the resulting predicted intensities with and without this assumption. Figure 10 shows a comparison of the predicted to the observed intensities from DEMs calculated with and without the /noblend option in aia_get_response. For the FAN1 DEMs, removing the blending improves the predicted intensity for the 335 and 131 Å channels. The 94 Å predicted intensity is somewhat worse for the FAN1 DEMs, although the error bars show that it is still a reasonable number. The ARC1 DEMs show improvement in the 335 Å channel without the blending, although the 94 Å channel is a little worse. The ARC2 DEMs show significant improvement in the predicted intensities in the 94 Å and 131 Å. The predicted intensities for the BG and SAD1 boxes are not affected significantly by the blending.

**Figure 10.** Comparison of predicted to observed intensities for DEMs calculated with and without blending of the 131 and 335 Å channels in the AIA response functions for the boxes indicated in Figure 7. Ratios of predicted to observed intensities from DEMs including the blending are shown in gray, those without it are shown in black.
Download figure:
Standard image High-resolution image

It should be noted that the DEMs themselves are extremely similar with and without incorporating the blending in the AIA response functions. Only the DEM for ARC2 shows any significant difference—without the blending the peak temperature shifts slightly higher. Because removing the blending on balance improves the predicted intensities, we use this option for all the DEMs calculated in this paper.

3.3. 2012 January 16 Flare

At 02:30 UT on 2012 January 16, a solar flare began on the Sun's northeast limb. It built up gradually and reached a magnitude of C6.5, with its peak occurring at about 04:44 UT. Many long, thin downflows occurred from 03:30 UT until 06:00 UT, by which time the flare had faded considerably.

For this flare, both XRT and AIA data are available. An AIA 131 Å image and an XRT Be-thin image are shown in Figure 11. Three boxes have been placed in the most noticeable downflows in the image. The box for SAD2 has been made narrower so that all pixels inside it are within the downflow (this condition was assured by plotting the 131 Å intensities of the pixels as a function of pixel number, and adjusting the box until it was clearly free of high intensity pixels from the surrounding arcade). The resolution of the 131 Å image has been degraded to match the XRT spatial resolution. Except for the SAD2 box, the boxes used for the DEMs are 7 × 7 XRT pixels. The XRT filters Be-thin, Al-med, and Be-thick were used in the analysis of this flare.

Figure 12 shows the DEMs calculated using the average intensities in all the boxes in the 2012 January 16 flare. The DEMs in Figure 12 have all been made by combining AIA and XRT data. It is apparent that SAD1 is slightly hotter than the surrounding supra-arcade plasma (FAN1), with a peak temperature of log(T) of 7.1 for SAD1 and log(T) of 7.0 for FAN1. The other two SAD DEMs peak at about the same temperature as their corresponding fan regions. The DEM for SAD2 has a slightly broader temperature distribution than the corresponding fan plasma.

The DEMs using both XRT and AIA data tend to have one or two more peaks than those with only AIA data, which is because we are using data for nine filters instead of six. The algorithm for calculating DEMs has more spline knots when more filters are used to create the DEMs (see Section 2.4). The more parameters that are included, the more complicated the final DEM can be, and therefore the more peaks will be present.

At a temperature of 100 MK, a large spike is visible in some of the DEMs, but there are several reasons for assuming that this feature is not representative of real temperatures in the observed plasma. AIA is incapable of detecting plasma at these temperatures, and temperature discrimination of the XRT temperature responses are not good enough to definitively identify plasma at this temperature, so the algorithm that calculates the DEMs tends to put more emission measure where there is a lack of constraints (Schmelz et al. 2009c). Additionally, these temperatures are physically highly improbable. For the DEMs that show this high temperature component, we have recalculated the DEMs with the constraint that the maximum log(T) = 7.5. These DEMs are very similar to the ones shown in Figure 12, with similar fits to the data, so we conclude that the high temperature spike is not an important component in the DEM.

Figure 12 also shows the predicted observations from the DEMs compared to the measured intensities. In general, the predicted intensities of the AIA filters agree quite well with the observations. The emission in the XRT Be-thin and Al-med filters are underpredicted by the FAN1 DEM, and the Be-thick filter for SAD2 is also underpredicted, although the error bars on the Be-thick signals are large because the signal in this filter is low, especially in the SAD regions.

The Monte Carlo runs for SAD1 and FAN1 are shown in Figure 13. The peak temperature is well defined in the Monte Carlo simulations for both DEMs. In fact, all of the Monte Carlo simulations for the FAN1 case peak at log(T) = 7.0, and 98% of the Monte Carlo simulations for SAD1 peak at log(T) = 7.1. The uncertainties at low and very high temperatures are large in the SAD1 DEM, which is due to the broadband nature of the XRT temperature response functions, as well as the low signal in the XRT filters in these regions. However, because the XRT filters can detect temperatures above log(T) = 7.5, the Monte Carlo solutions at these temperatures provide a useful constraint on the plasma DEM, where the Monte Carlo solutions for DEMs calculated with AIA alone do not. In the SAD1 case, more than 80% of the solutions show no DEM component at 7.5 ⩽ log(T) ⩾ 7.8 that is within two orders of magnitude of the peak of the DEM, indicating that a significant hot component to the plasma temperature profile is unlikely.

We calculate the DEMs using only the AIA data for this flare, and a comparison of the DEM of SAD1 with and without the inclusion of XRT data is shown in Figure 14. The effect of adding XRT data is to make the hot peak narrower, which is because this temperature range is well covered by XRT, but at the limit of the coverage of AIA. We also find that overall the predicted intensities from the DEMs including XRT data agree with measured observations better than DEMs without XRT data do. This comparison, as well as the fact that the XRT filter responses cover a higher temperature range than the AIA filter responses, implies that the DEMs that include XRT data display a more reliable picture of the temperature structure of a flare than DEMs calculated with AIA data alone.

**Figure 14.** Comparison of the DEMs calculated for SAD1 using AIA and XRT data (solid line) and using only AIA data (dashed line).
Download figure:
Standard image High-resolution image

3.3.1. The January 16 Flare at an Earlier Time

As the 2012 January 16 flare evolves, the AIA 131 Å channel shows a structure rising up in the line of sight of SAD1, appearing to fill up the downflow with hot plasma, beginning at about 05:05 UT. The image in Figure 11 was taken at 05:22:59. Because the DEMs give the temperature distribution of all the plasma along the line of sight, this plasma could be skewing the emission measure of the SAD, giving hotter temperatures than the surrounding plasma. Thus we examine this downflow at an earlier time (i.e., before this filling in began) in order to see if this structure is influencing the DEM.

We look at the SAD1 downflow at 04:48 UT. XRT was pointing just south of the flaring region, but SAD1 is still visible in the upper part of the field of view. XRT was not in flare mode at the time, so none of the thick filters are available, but observations were taken in the Ti-poly and Be-thin filters. Figure 15 shows an AIA 131 Å image and an XRT Be-thin image at this time. There is more contrast between SAD1 and the supra-arcade plasma than at the later time, and it is easy to make out the boundary between the downflow and the top of the arcade, unlike at the later time when the downflow appears to gradually fade into the top of the arcade.

**Figure 15.** AIA 131 Å (left) and XRT Thin-Be (right) images around 04:48 UT for the 2012 January 16 flare showing the locations of the regions used in the DEM calculations. The color table has been reversed for both images. Note that the AIA image has been degraded to match the XRT spatial resolution.
Download figure:
Standard image High-resolution image

Figure 16 shows the DEMs for this early time, as well as a comparison of the predicted and observed intensities. The emission measure is still lower for the SAD than for the fan plasma, but the peak temperatures for the two DEMs are the same. The predicted intensities for SAD1 and FAN1 concur very well with the observed intensities, with all of them being within a factor of 1.5 of the measured intensities, except for the Ti-poly filter in the SAD DEM, although the errors on the observed emission of this filter place it in range of the predicted intensity.

The DEMs for the SAD and the surrounding plasma at the early and late times have each been combined in one plot in Figure 17 in order to show the evolution of the DEMs. We find that the downflow at the early time peaks at log(T) = 7.0 and at the late time peaks at log(T) = 7.1, so the plasma in the SAD appears to be heating slightly with time. Additionally, the emission measure at the peak temperature increases from the early to late times, confirming the impression from the images that the SAD is filling up with hot plasma, as observed in the 131 Å images. The DEMs for the surrounding fan plasma are quite similar at the early and late times, peaking at log(T) = 7.0 at both times. The emission measure at log(T) = 7.0 at the late time is slightly higher than the emission measure at the same temperature at the early time. The filling effect we observe in the 131 Å images could be due to hot plasma in the temperature range imaged by the 131 Å channel that is evaporating up into the region of the current sheet as the reconnection progresses, as has been seen in models (e.g., Reeves et al. 2010; Reeves & Golub 2011).

3.4. 2012 January 27 Flare

The flare that took place on 2012 January 27 on the Sun's northwest limb was extremely bright, and was accompanied by a coronal mass ejection. It started at 17:30 UT and reached a magnitude of X1.7. A large number of long, thin SADs could be observed from around 19:00 (soon after the peak flare intensity) until the early hours of the next day when the flare eventually died down. The clear structure of the plasma loops that constitute the arcade make it easy to infer that there is a huge accumulation of magnetic energy present. Because XRT was pointing at the whole flare structure for a long period of time, it was studied at two times: 19:53 UT and 20:45 UT.

3.4.1. 2012 January 27 at 19:53 UT

Figure 18 shows an AIA 131 Å image and an XRT Be-thin image for the 2012 January 27 flare at the earlier time of 19:53 UT, with the boxes used for DEMs. Several downflows are visible here, but the two judged to be the most prominent were selected for study, labeled SAD1 and SAD2. The boxes for SAD1 and SAD2 have been narrowed in order to accommodate the slender shapes of the downflows in this flare. The same method is used as in the narrowing of the SAD2 box in Section 3.3 to be sure that no fan plasma was incorporated in the SAD boxes. A movie in 131 Å is used to spot these most prominent downflows at this stage in the flare's development. The SAD marked SAD2 moves extremely fast and disappears into the flare arcade in a matter of a couple of minutes. The images in Figure 18 catch this SAD as it first moves through the fan plasma.

The DEMs for the SADs and fan plasma are shown in Figure 19. The XRT filters used at this time were Be-thin, Be-med, Be-thick, and Al-thick. The DEM for FAN1 peaks at a slightly higher temperature log(T) = 7.1) than the DEM for SAD1 (log(T) = 6.9). The DEM for FAN2 peaks at log(T) = 6.3, whereas the DEM for SAD2 peaks at log(T) = 6.5, but the temperature distribution for the plasma in FAN2 is fairly broad, and there is a large amount of plasma at and above the peak temperature for the SAD2 DEM. Additionally, examining the Monte Carlo simulations for this case shows that the peak temperature for the FAN2 distribution is not well defined. For SAD2, 75% of the Monte Carlo simulations have a peak in the DEM at log(t) = 6.5, whereas 23% have a peak temperature lower than log(T) = 6.5. For FAN1, 45% of the Monte Carlo simulations have a peak at logt(T) = 6.5 or above. Because of the broadness in the FAN2 DEM, and the uncertainty in its peak value, it would not be correct to say that the SAD plasma is hotter than the fan plasma in this case. However, it is clear that the plasma in this region is overall cooler than the plasma in the region of SAD1 and FAN1.

The comparisons between the predicted and observed intensities for the DEMs are also shown in Figure 19. For SAD1 and FAN1, the model observations match up well with the measured intensities, although the Be-thin intensities are overpredicted and the Be-thick intensities are underpredicted for SAD1. As in the 16 January 05:22 UT case, the error bars on the Be-thick filter are large because the counts are low. SAD2 and FAN2 have reasonable matches between the predicted and observed intensities, but the AIA 94 Å filter is not well predicted by either DEM, and the XRT Be-thick filter is underpredicted (but within errors) in both cases. One possible explanation for the discrepancies in this event is that the DEM calculations implicitly assume that the plasma is in thermal equilibrium. This SAD moves very quickly, and dynamically evolving, relatively low-density plasma can be affected by nonequilibrium ionization effects. Modeling efforts by Shen et al. (2013) have recently shown that nonequilibrium ionization effects can influence the observed intensities in AIA telescopes in structures surrounding the current sheet in a solar eruption.

3.4.2. 2012 January 27 at 20:45 UT

Almost an hour later, the structure of the flare arcade was largely unchanged. Figure 20 shows an AIA 131 Å image and an XRT Be-thin image at 20:45 with different downflows. The gradual sunward movement of downflows over time and a more sharply defined supra-arcade region make it possible to choose box locations closer to the top of the arcade. It should be pointed out that these are different downflows than those studied at the earlier time. The downflows at this time in the flare more closely resemble the dart-like shapes of the models by Cécere et al. (2012). Since one pixel in XRT is ∼1 arcsec and 1 arcsec on the Sun is about 725 km, the downflows in this image are about 3000 km wide and about 15,000 km long, which means their sizes are also similar to the sizes of the models. Boxes in the downflows have been made narrower to accommodate the shapes of the downflows. Be-thin, Al-med, and Be-thick have been used as the filters on XRT.

Figure 21 shows the DEMs for the boxes shown in Figure 20. Both SAD1 and SAD2 peak at lower temperatures than the arcade plasma surrounding them. In fact, the SAD1 DEM does not show any evidence of a high temperature peak at all, and the background (BG) DEM actually peaks higher than the SAD1 DEM. SAD2 peaks at log(T) = 6.5, and the corresponding fan plasma DEM peaks at log(T) = 6.6. As with the earlier time, the plasma in the region of SAD2 and FAN 2 is overall cooler than the region of SAD1 and FAN1.

Figure 21 also shows the comparison between the intensities predicted by the DEMs and the observed intensities. The agreement is quite good for SAD1 and FAN1. The agreement is not as good for FAN2 and SAD2, with the AIA 335 Å filter being overpredicted for each structure, and the AIA 94 Å underpredicted for SAD2. Although the plasma in this region is not quite as dynamic as at the previous time, nonequilibrium ionization effects could still be playing a role here. For all these DEMs, the XRT Be-thick is underpredicted, as they are for the DEMs calculated for the earlier time.

Comparing the two times that we studied for this flare, it appears that there is more hot plasma in SAD1 at the earlier time (although the SADs are not hotter than the supra-arcade plasma at either time in this region). However, the fan plasma in the area of FAN1 stays relatively constant, indicating that over time, SADs have a tendency to cool down in this region. The characteristics of the DEMs in the region of SAD2 and FAN2 are fairly similar from the earlier time to the later time, even though the specific SADs they are representing are different. Because the DEMs for fan plasma in other events have a tendency to be fairly similar over the timeframe in question, it is possible that there is some fan plasma in the line of sight that is dominating the DEM for SAD2, both at the early and late times.

4. DISCUSSION AND CONCLUSIONS

In total, we have examined 11 SADs from four separate flares, and one of the downflows has been examined at two separate times. We find that only one of these downflows (SAD1 in the 2012 January 16 flare) definitively indicates temperatures slightly hotter than those of the surrounding arcade. We find that another downflow (SAD2 from the 2012 January 27 flare at 19:53 UT) has a higher peak temperature than the surrounding fan plasma, but the temperature distribution of the fan plasma is broad, and its peak temperature is uncertain. We find that the other two downflows from the 2012 January 16 flare show peak temperatures roughly the same as the surrounding arcade, and the remaining downflows studied have temperatures slightly cooler than the surrounding arcade. Higher temperatures and emission measures were found in the downflows compared to the background areas in most cases.

Very few models of SADs predict the temperatures of these structures, with the models of Maglione et al. (2011) and Cécere et al. (2012) being the only models we are aware of that present predictions for SAD temperatures. The downflows in the Maglione et al. (2011) model exhibit temperatures of 84 MK inside the downflow and 5.7 MK outside, whereas the Cécere et al. (2012) models have temperatures ranging from 22 MK to 29 MK inside the downflows to 0.56 MK outside. In these models, thermal conduction is neglected. Both the temperature and temperature gradient in the simulations are large, so the fact that their models omit conduction suggests that their resulting temperatures could be overestimates. Nevertheless, these models predict heating within the supra-arcade voids, leading to elevated temperatures compared to the surrounding plasma.

In our observations, however, no signs of highly elevated temperatures in SADs are found. The SAD temperatures we find are very similar to those of the arcade and their emission measures are lower, typically by an order of magnitude. In the models discussed above, the densities inside the downflows are about an order of magnitude less than the surrounding plasma, which would lead to an emission measure that is two orders of magnitude less. Since the DEMs are the temperature distribution for all the plasma along the line of sight, it is possible that a weaker component due to plasma inside the SADs is masked by a stronger component due to some fan plasma in front of or behind the SAD. Our data suggests that there is a similar distribution of plasma temperatures in the downflows and supra-arcade region (although in some cases a cooler distribution), and that the plasma in the SADs is less dense. In most of the cases we examined, a weak component at the same temperature as the fan plasma could be masked by surrounding fan plasma, but fan plasma in the line of sight would not mask a weak component to the DEM with significantly higher temperatures than the fan plasma. Because we do not observe such a weak, high-temperature component in the SAD DEMs, we conclude that there is no evidence that SADs contain substantially hotter plasma than in the surrounding arcade. There is little convincing evidence that temperatures equal to or greater than 22 MK are present in the SADs.

Flares are typically examined about an hour after the peak flare intensity, which is when most downflows are present. When analyzed at two different times, some the SADs in the 2012 January 27 flare seemed to have cooled down, indicating there is a possibility that the Maglione et al. (2011) and Cécere et al. (2012) models could be correct if the DEM analysis was done at the immediate creation of the SADs. However, since the three downflows in the 2011 October 22 flare appear in sequence and travel at similar velocities, it would be reasonable to assume the downflows closest to the solar surface are minutes older. In this flare, the older the downflow, the hotter it is, which is most likely because the surrounding plasma is hotter. So the most recently created SAD in this flare does not show evidence of hot plasma. In addition, the SAD2 downflow in the 2012 January 27 flare at 19:53 UT appeared seconds before the image was taken. Looking at its DEM, there is no evidence to suggest extremely high temperatures in this downflow, and it even shows temperatures significantly lower than those in other parts of the arcade.

Elevated temperatures in SADs and the large temperature differences between the SADs and the surrounding plasma in the models by Maglione et al. (2011) and Cécere et al. (2012) are not supported by the data. Therefore, it seems likely that some other mechanism must be responsible for creating SADs. One possibility is the observational interpretation presented by Savage et al. (2012), where the SADs are voids cleared of plasma by descending flux tubes. This scenario has recently been modeled by Scott et al. (2013), and they find that a descending flux tube leads to a shock in front of the flux tube, with a rarefaction region behind it, causing the SAD. Another possibility is the model by Cassak et al. (2013), which posits that the SADs are flow channels from reconnection outflows in a density stratified atmosphere. The implications of both of these models for the temperatures inside the SADs have yet to be worked out in detail, but it seems likely that they would both predict lower temperatures inside the SADs when compared to the supra-arcade fan plasma (R. Scott & P. Cassak 2013, private communication).

Possible future work includes expanding the analysis done here to more events. Several dozen flares with SADs have been discovered since 2010 February using AIA. A more reliable test against current models of SADs can be achieved by creating a larger wealth of data using AIA and XRT observations of these flares. Further study into the evolution of the temperatures in the SADs is also a possible interesting avenue of research, and could be useful for understanding which of the proposed mechanisms is most likely to be causing the SADs. However, further modeling work is also needed in order to understand the implications of the results presented here, since the detailed thermodynamics of the competing proposed mechanisms have yet to be worked out.

The authors would like to thank the anonymous referee for helpful comments that improved this paper, and P. Jibben, S. Saar, and M. Weber for help with the XRT point-spread function. This paper was part of a master's thesis from the University of Southampton. Will Hanneman would like to thank J. Drake and P. Charles for maintaining the connection between the University of Southampton and the Harvard-Smithsonian Center for Astrophysics. Travel to the AAS meeting and part of K.K. Reeves contribution supported under contract SP02H1701R from Lockheed-Martin to SAO. K. K. Reeves is also supported under contract NNM07AB07C from NASA to SAO and NASA grant NNX13AG54G. Hinode is a Japanese mission developed and launched by ISAS/JAXA, with NAOJ as domestic partner and NASA and STFC (UK) as international partners. It is operated by these agencies in co-operation with ESA and the NSC (Norway). This work has benefited from the use of NASA's Astrophysics Data System.

Facilities: Hinode (XRT) - Hinode (Solar-B) , SDO (AIA) - Solar Dynamics Observatory

THERMAL STRUCTURE OF CURRENT SHEETS AND SUPRA-ARCADE DOWNFLOWS IN THE SOLAR CORONA

Article metrics

Permissions

Author e-mails

Author affiliations

Dates

ABSTRACT

1. INTRODUCTION