1: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2: %

3: % final version submitted to ApJ

4: % with changes in response to the referee's comments

5: %

6: % Paolo Grigis, Institute of Astronomy, ETH, Zurich (2005-2007)

7: % Paolo Grigis, Center for Astrophysics, Cambridge MA (2007-2008)

8: % pgrigis@astro.phys.ethz.ch

9: % 

10: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

11: %

12: %

13: %

14: % Paolo C. Grigis and Arnold O. Benz

15: % Spectral Hardening in Large Solar Flares

16: % 

17: % 

18: 

19: % packages etc.

20: %\documentclass{aastex}

21: %\documentclass[12pt,preprint]{aastex}

22: %\documentclass[preprint2]{aastex}

23: 

24: \documentclass{emulateapj}%when uncommenting this, replace %\clearpage in figs

25:                           %with \clearpage (!)

26: 

27: %\usepackage{txfonts}

28: %\usepackage{graphicx} 

29: 

30: %this enables ok printing on A4 page

31: \setlength{\voffset}{-2.5cm}

32: 

33: % commands definition

34: \newcommand{\diff}[2]{\frac{\partial #1}{\partial #2}}

35: \newcommand{\difff}[2]{\frac{\partial^2 #1}{\partial #2^2}}

36: \newcommand*{\emes}{\mathcal{M}}    % Emission Measure

37: \newcommand*{\eturn}{{E_\mathrm{turn}}} % Turnover Energy

38: \newcommand*{\eref}{{E_0}} % Normalization (reference) Energy

39: \newcommand*{\ftot}{F_\mathrm{tot}}%Total non-thermal flux

40: \newcommand*{\flux}[1]{F_{\!{#1}}}%Flux at #1 keV

41: \newcommand*{\estar}{E_*}

42: \newcommand*{\fstar}{F_*}

43: \newcommand*{\epstar}{\epsilon_*}

44: \newcommand*{\phistar}{\Phi_*}

45: \newcommand*{\eps}{\epsilon}

46: \newcommand{\avck}{\langle k\rangle}

47: \newcommand{\ele}{E}

48: \newcommand{\phe}{\varepsilon}

49: \newcommand*{\etal}{et al.\ }

50: % end commands

51: 

52: \shorttitle{Spectral Hardening in Large Solar Flares}

53: 

54: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

55: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

56: %

57: \begin{document}

58: %

59: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

60: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

61: 

62: 

63: \title{Spectral Hardening in Large Solar Flares}

64: 

65: 

66: \author{P. C. Grigis\altaffilmark{1,2}}

67: \affil{pgrigis@cfa.harvard.edu}

68: \and 

69: \author{A. O. Benz\altaffilmark{1}}

70: \affil{benz@astro.phys.ethz.ch}

71: \altaffiltext{1}{ETH Zurich, 8092 Zurich, Switzerland}

72: \altaffiltext{2}{Harvard-Smithsonian Center for Astrophysics, Cambridge MA 02138, USA}

73: 

74: 

75: \begin{abstract}

76: 

77:   Observations by the Ramaty High Energy Solar Spectroscopic Imager (RHESSI) are

78:   used to quantitatively study the hard X-ray evolution in 5 large solar flares

79:   selected for spectral hardening in the course of the event. The X-ray

80:   bremsstrahlung emission from non-thermal electrons is characterized by two

81:   spectroscopically distinct phases: impulsive and gradual. The impulsive phase

82:   usually consists of several emission spikes following a soft-hard-soft

83:   spectral pattern, whereas the gradual stage manifests itself as spectral

84:   hardening while the flux slowly decreases.

85: 

86:   Both the soft-hard-soft (impulsive) phase and the hardening (gradual) phase

87:   are well described by piecewise linear dependence of the photon spectral index

88:   on the logarithm of the hard X-ray flux. The different linear parts of this

89:   relation correspond to different rise and decay phases of emission spikes.

90:   The temporal evolution of the spectra is compared with the configuration and

91:   motion of the hard X-ray sources in RHESSI images.

92: 

93:   These observations reveal that the two stages of electron acceleration causing

94:   these two different behaviors are closely related in space and time. The

95:   transition between the impulsive and gradual phase is found to be smooth and

96:   progressive rather than abrupt. This suggests that they arise because of a

97:   slow change in a common accelerator rather than being caused by two

98:   independent and distinct acceleration processes. We propose that the hardening

99:   during the decay phase is caused by continuing particle acceleration with

100:   longer trapping in the accelerator before escape.

101: 

102: \end{abstract}

103: 

104: 

105: \keywords{Sun: flares -- Sun: X-rays, gamma rays -- Acceleration of particles}

106: 

107: 

108: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

109: %

110: \section{Introduction}

111: %

112: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

113: 

114: 

115: Large solar flares are very bright hard X-ray sources. The emission originates

116: from energetic electrons with energies mainly in the 10s and 100s of keV,

117: believed to be accelerated in the corona. These electrons have a short lifetime

118: in regions dense enough to generate substantial hard X-ray emission, and

119: therefore react quickly to changes in the acceleration, transport and emission

120: processes. While it may be hard do disentangle the contributions of the various

121: effects in different flares, the investigation of the temporal evolution

122: of the hard X-ray spectra in single flares is a valuable tool to study these

123: processes. Flare models and theories should be able to account for the behavior

124: of the observed hard X-ray spectra as they change during an event.

125: 

126: Observations of the spectral hard X-ray evolution have revealed two main

127: trends: a soft-hard-soft (SHS) spectral evolution of emission peaks, and a

128: progressive hardening during whole events (SHH, soft-hard-harder).

129: 

130: The SHS behavior of emission spikes was discovered by \cite{parks69}, and since

131: then has been reported by many others. Recently, \cite{grigis04} surveyed

132: quantitatively the spectral evolution of emission spikes during M class events,

133: finding that nearly all rise and decay phases of the peaks show the SHS

134: behavior. The excursions in both photon flux measured at a fixed energy and

135: spectral hardness can be very different from peak to peak. However, they show

136: consistently a characteristic property: the spectral power-law index is a linear

137: function of the logarithm of the flux \citep{grigis05}. The SHS pattern has also

138: been observed in looptop sources \citep{battaglia06}. Thus it is likely to be a

139: characteristic signature of acceleration rather than a propagation

140: process. Detailed modeling of transit-time damping acceleration of electrons can

141: reproduce the SHS behavior if the effects of particle trapping in the

142: acceleration region and escape are taken in account \citep{grigis06}.

143: 

144: The SHH behavior was first observed by \citet{frost71}, who noted that the

145: spectral index in the late phase of a flare stayed constant at a harder value

146: than measured during the first (impulsive) SHS peak. Further events were studied

147: by \citet{cliver86} and \citet{kiplinger95} using data from the Hard X-Ray Burst

148: Spectrometer (HXRBS) on SMM. The distinctive feature of the SHH evolution is the

149: absence of softening as the flux decreases.

150: 

151: Kiplinger found two different subtypes of behavior:

152: %

153: \begin{itemize}

154: \item hardening during a particular peak

155: \item hardening during the decay of the whole event

156: \end{itemize}

157: %

158: In the first subtype, substantial hardening occurs during a short period, but

159: after an emission peak the flux may soften again. Events of the second subtype

160: (corresponding to the classic flare with a gradual phase) typically have some

161: SHS peaks at the beginning but progressively harden afterwards. Despite the name

162: SHH, a general hardening may start already before the largest peak. Thus SHH is

163: not limited to the decay phase. We note also that the two classes of spectral

164: evolution are not clearly separated: most SHH events show impulsive SHS peaks in

165: the beginning. The interest in SHH flares rose after Kiplinger's report of a

166: high association rate between SHH and the occurrence of interplanetary energetic

167: proton events. More recently, \citet{2008ApJ...673.1169S} confirmed the

168: association between solar energetic particles and hardenings during the January

169: 2005 solar storm events. The link between the two phenomena, however, is 

170: not the subject of this work.

171: 

172: The two different kinds of spectral evolution (SHS and SHH) seem to support the

173: view that there are two different stages in flares: an impulsive phase at the

174: beginning followed by a gradual component, corresponding to different

175: acceleration mechanisms. This scenario was first proposed to explain radio

176: observations \citep{wild63}. A first (impulsive) phase was suggested to

177: accelerate electrons producing gyrosynchroton emission, and the second (gradual)

178: phase was linked to traveling shocks (type II radio bursts) accelerating further

179: the electrons and also ions. This idea was then used to interpret hard X-ray

180: observations \citep{frost71, bai1979}. Later, the shocks were associated with Coronal Mass

181: Ejections (CMEs). Occulted flares seemed to confirm this scenario

182: \citep{hudson82}. However, it cannot explain the position of the dominant hard

183: X-ray source seen during the gradual phase: imaging observations by Hinotori

184: \citep{ohki83} showed that the hard X-ray emission comes from too low in the

185: solar corona to justify the connection with type II radio

186: bursts. \citet{kahler84} and  \citet{bai1986} argued that the impulsive phase is followed by two

187: independent acceleration processes. The first happening in the post-flare loop

188: arcade is responsible for the late-phase hard X-ray emitting electrons, and the

189: second higher up in the corona, shock driven, accelerates interplanetary

190: electrons and ions. This was later corroborated by \citet{cliver86} using SMM

191: observations. 

192: 

193: Stochastic acceleration reproduces the observed SHS behavior but 

194: cannot at the same time describe hardening when the flux decays. If both SHS and

195: SHH phases of electron acceleration happen in the same event, why does the

196: spectral behavior reverse? Are different acceleration mechanisms at work, or is

197: there a further parameter in the same process that changes in the course of the

198: flare? To find an observational answer to this question, simultaneous imaging

199: and spectral observations are analyzed that have become available for the first

200: time by the Reuven Ramaty High Energy Solar Spectrometric Imager

201: \citep[RHESSI;][]{lin02}. Its data characterize the spectral evolution of the

202: non-thermal hard X-ray flux in unprecedented detail. We also compare the

203: spectral evolution with the geometrical flare configuration, as well as the

204: motions of the coronal X-ray source and the non-thermal footpoint sources.

205: 

206: 

207: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

208: %

209: %TABLE 1: Event description

210: %

211:  

212: \begin{table}

213: \caption{Chronological list of the selected events.}

214: \label{tab:evlist2}

215: \centering

216: \begin{tabular}{r c c l}

217: \hline\hline

218: \multicolumn{1}{c}{Date}

219: & \multicolumn{2}{c}{\hspace{0.3cm}GOES peak\hspace{0.6cm} } & Panel in \\

220:       & time & flux  & Figs. \ref{fig:lightcurves} and \ref{fig:glogf}\\

221: \hline

222:   7-NOV-2004 & 16:05 & X2.0 & F\\ 

223:  10-NOV-2004 & 02:13 & X2.6 & A\\ 

224:  17-JAN-2005 & 09:52 & X3.9 & B\\ 

225: 

226:  19-JAN-2005 & 08:23 & X1.4 & C \& D\\ 

227:  20-JAN-2005 & 07:01 & X7.1 & E\\ 

228: \end{tabular}

229: \end{table}

230:  

231: %

232: %

233: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

234: 

235: 

236: 

237: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

238: %

239: \section{Method}

240: \label{methods}

241: 

242: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

243: %

244: %Figure 1: spectrum & fitted model

245: %

246:  

247: \begin{figure}

248: %\resizebox{\hsize}{!}{\includegraphics{fig1.eps}}

249: \plotone{f1.eps}

250: %

251: \caption{

252:   Example of count spectrum ({\it thick error bars}) observed by RHESSI,

253:   integrated for 4 seconds around 17-JAN-2005 09:43:36 UT. The spectrum was

254:   fitted with a thermal component (Maxwellian), a non-thermal component

255:   (log-parabolic, see Appendix A), and the average spectrum before and after the

256:   flare, assumed to be the background. For clarity, the total model consisting

257:   of the sum of the three components is not plotted over the observed count

258:   spectra, but its normalized residuals are shown below instead. The residuals

259:   show that the fitted spectral model reproduces the observed counts. The reduced

260:   $\chi^2$ for this spectrum equals 0.94.   }

261: \label{fig:countspectrum}

262: \end{figure}

263: 

264: %

265: %

266: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

267: 

268: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

269: %

270: %Figure 2: results

271: %

272: \begin{figure*}

273: %\epsscale{1.0}

274: \plotone{f2.eps}

275: %

276: \caption{

277:   Temporal evolutions of the spectral index $\gamma$, the flux at 50 keV,

278:   $I_{50}$, and the spectral curvature $\eta$ for the selected events. 

279:   Below, the correlation coefficient between

280:   the two curves is shown vs. time. It was determined during the time intervals

281:   indicated by the horizontal bars. The vertical bars represent the 68\%

282:   confidence range.  The onset times of the flare associated CMEs are marked by

283:   arrows.  O1 and O2 indicate linear and quadratic extrapolations, respectively,

284:   to the CME altitude vs. time evolution.}

285: \label{fig:lightcurves}

286: \end{figure*}

287: %

288: %

289: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

290: 

291: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

292: %

293: %Figure 3: results, in gamma-log F space

294: %

295: \begin{figure*}

296: \epsscale{0.6}

297: \plotone{f3.eps}

298: %

299: \caption{Evolution of non-thermal X-ray spectral index $\gamma$ vs. photon flux

300:   $I_{50}$ in the course of flares.}

301: %

302: \label{fig:glogf}

303: \end{figure*}

304:  

305: %

306: %

307: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

308: 

309: The goal of this paper is a detailed quantitative study of the spectral

310: evolution of solar flares showing a hardening trend in RHESSI observations.

311: Rather than attempting a statistical study of a large number of flares, the

312: analysis is restricted to a few events studied exhaustively. Therefore, we do

313: not estimate the occurrence frequency of SHH flares or their rate of association

314: with solar energetic particle events. This has been done by \cite{kiplinger95}

315: using SMM/HXRBS data, who reports 24 occurrences of hardening out of 152 events

316: with peak flux count rate larger than 5000 counts s$^{-1}$. Most of the SHH

317: events reported by Kiplinger are in the upper M and X GOES class. The reality of

318: this trend needs however to be confirmed.

319: 

320: We selected flares with a GOES flux above X1 during RHESSI observation time

321: windows. 50 events satisfying this condition were found in the period from the

322: start of the mission (February 2002) to September 2006. We additionally required

323: that the rise, main and decay phases were well observed to study the spectral

324: evolution in time. This left us with 12 candidates.

325: 

326: As a first approximation, the presence of hardening behavior was tested by

327: studying the count rates in the energy range from 30 to 60 keV. We fitted the

328: spectral index separately in three energy bands (30--40 keV, 40--50 keV, 50--60

329: keV) and looked for either trends of progressive hardening or the presence of a

330: late hard phase. The lowest of these bands is sometimes contaminated by thermal

331: emission. This can be easily spotted by comparing the time profile with the

332: other bands. Using count rates is adequate to identify candidate events for

333: spectral hardening, but may have missed some events where hardening happens in a

334: phase of low flux close to the background. After discarding two further events

335: with high pileup, we found 5 well-observed events with a clear signature of

336: hardening. These are listed in Table \ref{tab:evlist2}.

337: 

338: For each of the selected events, the instrumental response matrices and

339: count-rate spectrograms covering the energy range from 3 to 500 keV were

340: generated for the front segments with a temporal resolution of one RHESSI spin

341: period (approximately 4 s). The spatially integrated photon spectra in the range

342: 12 to 500 keV were fitted with two components: an isothermal component at low

343: energies (below about 20-40 keV) and a non-thermal component at higher

344: energies. The shape of the non-thermal component is assumed to be a

345: log-parabolic curve characterized by three parameters: its normalization

346: $I_{\eps_0}$ at energy $\eps_0$, its spectral index $\gamma <0$ and its spectral

347: curvature $\eta$.  The functional form of this model (Eq. \ref{eq:spmodel}), as

348: well as the rationale for choosing it are explained in detail in

349: App. \ref{app:a}. Figure \ref{fig:countspectrum} shows an example of an observed

350: count spectrum and the best-fit components. The normalized residuals

351: indicate an excellent fit.

352: 

353: During the fitting process, the model photon spectrum is folded with the

354: response matrix, yielding the expected count spectrum from the model. The

355: background spectrum is then multiplied with a normalization factor $\lambda$ and

356: added to the model counts, where $\lambda$ is an additional fit parameter for

357: the model fitting and is constrained between 0.5 and 2. This correspond

358: approximately to the maximum excursion in RHESSI's background in the front

359: segments during an orbit. The parameter $\eta$ is constrained to be zero or

360: positive (corresponding to a parabola in log-log bending down). This ensures

361: that the emission approaches 0 for infinitely high energies.

362: 

363: The large amount of data (more than 3 thousand spectra) required an automated

364: fitting routine. For every spectrum, 2 preliminary passes were done estimating

365: the parameters for the thermal and the non-thermal part which were then used as

366: starting parameters for the final fitting. This turned out to deliver good

367: fittings for most of the data. A check of the quality of the fits was performed

368: by looking at the time evolution of $\chi^2$ and of the fitting

369: parameters. Spectra with reduced $\chi^2$ worse than 2 were manually fitted

370: again, and in most cases it was possible to find another set of fit parameters

371: yielding reduced $\chi^2$ below 2, with the exception of the event of

372: 20-JAN-2005.

373: 

374: This event (the largest flare, GOES class X7) is characterized by very strong

375: thermal emission. At times when the non-thermal emission is weak and/or soft,

376: pileup effects are especially large in the 20-50 keV band. Therefore, the

377: fittings, which are good above 50 keV, have large residuals below that

378: energy. This may be due to the fact that the photon spectrum model chosen is not

379: suited to describe the observed photon spectra, or that the pileup correction is

380: inaccurate. Because it is very hard to correctly take into account pileup

381: effects in such a regime, it is not clear whether the model failure is real or

382: instrumental. Therefore, we let the spectrum model stand as it is, but caution

383: that the parameter values fitted in the flare of 20-JAN-2005 may be

384: less accurate, due to the unknown systematic effects generated by imperfect

385: pileup correction.

386: 

387: For the other events, the fit parameters are of good quality and the

388: corresponding photon models are a high-fidelity representation of the incoming

389: photon flux. The final distribution of the reduced $\chi^2$ for all events

390: (except 20-JAN-2007) is such that 89\% of all spectra have $\chi^2$ less than

391: 1.5 and 97\% of all spectra have $\chi^2$ less than 2. Therefore the unusual

392: choice of the logarithmic parabolic fit-model, explained in Sect. \ref{methods},

393: produces good fittings and is therefore justified \emph{a posteriori}.

394: 

395: Three of the events presented here have also been studied by

396: \cite{2008ApJ...673.1169S}, where the non-thermal component has been fitted by a

397: double power-law. The temporal evolution of our values for $\gamma$ (at 50 keV)

398: is very similar to theirs, although the actual numerical values are slightly

399: different due to the different fitting methods employed and their choice of

400: energy intervals. The differences are larger for the January 20 event, where

401: they used the 100-200 keV energy interval.

402: 

403: The thermal evolution of the events shows a rapid increase of the emission

404: measure at the beginning of the event, followed by a flat peak and a slow

405: decay. The temperature is in the range 20-40 MK, peaks before the emission

406: measure and decays faster.

407: 

408: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

409: 

410: 

411: \section{Spectroscopy Results}

412: \label{results}

413: 

414: Figure \ref{fig:lightcurves} shows time profiles of the photon spectral

415: index~$\gamma$ and the flux normalization at 50 keV,~$I_{50}$ (given in

416: Eq.~\ref{eq:spmodel}), for the events of Table~\ref{tab:evlist2}, as found by the

417: spectral fitting procedure explained in Sect.~\ref{methods} and Appendix A.

418: 

419: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

420: %

421: %Figure 4: histogram of distributions

422: %

423:  

424: \begin{figure}

425: %\resizebox{\hsize}{!}{\includegraphics{fig5.eps}}

426: \plotone{f4.eps}

427: %

428: \caption{Distribution of the slope in $\gamma$ vs. log($I_{50}$)

429:   (Fig. \ref{fig:glogf}, parameter $a$ in Eqs. \ref{eq:lintrendglogf} and

430:   \ref{eq:pivslope}) for the rise and decay phases in all peaks of the

431:   observed events (upper panel) and for rise and decay phases separately (middle

432:   and bottom panel, respectively).}

433: \label{fig:histoslope}

434: \end{figure}

435:  

436: %

437: %

438: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

439: 

440: The observed spectral variability of flares on time scales down to ten seconds

441: or less requires the highest possible temporal resolution for the spectral

442: fitting (about 4 seconds in our case). Longer integration times, while desirable

443: for better photon statistics, are not suitable because the averaging effect of

444: summing spectra with different hardness blurs the spectral evolution.

445: Nevertheless, during the decay phase the flux is so low that full resolution

446: spectra deliver noisy values for the fitting parameters. In this case, longer

447: integration times must be used. This is acceptable, since the variations of the

448: hard X-ray flux are slower during the decay phase, and short-lived spikes are

449: less frequent. Therefore, some light curves shown in Fig. \ref{fig:lightcurves}

450: and subsequent figures use a lower cadence of approximately 32 seconds (that is,

451: 8 RHESSI rotations) in the decay phase.

452: 

453: Short gaps lasting about 1 minute can be seen in the light curves

454: (Fig. \ref{fig:lightcurves}). They correspond to periods where the thick

455: attenuator was removed from the field of view, but the X-ray flux was still so

456: large that the dead time in the detector prevents meaningful spectral analysis.

457: 

458: The selected flares show many distinct SHS peaks. They are characterized by a

459: temporal correlation of $\gamma$ and $\log I_{50}$, yielding roughly parallel

460: curves in Fig. \ref{fig:lightcurves}. On the other hand, in the presence of SHH

461: peaks or progressive hardenings, the two time profiles diverge. This may be

462: illustrated in the event of 19-JAN-2005 (Fig. \ref{fig:lightcurves}, panel C),

463: where the two profiles run roughly parallel until 08:29 and then start to

464: diverge.

465: 

466: To better distinguish between the SHS and the hardening trends,

467: Fig.~\ref{fig:lightcurves} also shows the correlation coefficients between the

468: spectral index $\gamma$ and the logarithm of the flux $\log I_{50}$ vs.

469: time. The vertical bars represent the 68\% confidence range (corresponding to

470: one standard deviation). SHS peaks are characterized by a correlation

471: coefficient close to +1. Times, on the other hand, when the spectrum hardens

472: while the flux becomes lower have a negative value of the correlation

473: coefficient. We note that during the SHS times correlation is rather constant

474: (near +1), whereas during the hardening phase the correlation coefficients are

475: more erratic and often not close to -1. This indicates that it may not be

476: possible to find a behavior similarly well-defined for the periods of hardening

477: as it is found for the SHS peaks. In some cases (19-JAN-2005, 08:28 to 08:30),

478: the spectral hardness stays nearly constant while the flux decays. This yields a

479: correlation coefficient near zero.

480: 

481: It should be noted here that most periods showing hardening have low

482: flux. Therefore the corresponding time profiles may be more noisy, weakening the

483: correlation. Some of this effect was compensated by increasing the time interval

484: for correlation. Another effect is loss of correlation during a broad

485: peak. Again this can be taken into account by increasing the correlation

486: interval.

487: 

488: During the late phase of the event of 19-JAN-2005, in the RHESSI orbit following

489: the one featuring the main peak, an uninterrupted phase of hardening is seen.

490: From 09:35 to 10:00 the flux decays exponentially (see

491: Fig. \ref{fig:lightcurves}, panel D). After that time, the emission reaches a

492: hardness comparable with the one of the background, and it becomes impossible to

493: disentangle the two components by purely spectral methods. This event will be

494: investigated in more detail in Section \ref{modeling}.

495: 

496: We also compared the start of the hardening with the onset time of

497: flare-associated CMEs, taken from the SOHO LASCO CME catalog

498: \citep{2004JGRA..10907105Y}. In three cases (panel A, B, and F in

499: Fig. \ref{fig:lightcurves}), the onset of the CME precedes the start of the

500: hardening by 3 to 5 minutes, in one case by 15--20 minutes (panel C) and in one

501: case by 50 minutes (panel E). In all five events an associated CME was

502: present, but the hardening phase never starts before the CME onset.

503: 

504: Figure \ref{fig:glogf} shows the relationship between $\gamma$ and $\log

505: I_{50}$. Since the flux increases from left to right and the hardness increases

506: from top to bottom, SHS peaks show as piecewise linear trends with a negative

507: slope \citep{grigis04,grigis05}. On the other hand, progressive hardening during

508: flux decay is visible as a trend with positive slope. Such relations can

509: be written as

510: \begin{equation}

511: \label{eq:lintrendglogf}

512:   \gamma=-a\log(I_{\eps_0})+b\,,

513: \end{equation}

514: where $a<0$ for the SHS peaks and $a>0$ during hardening phases. The minus

515: sign in front of $a$ takes care of the fact that in Fig. \ref{fig:glogf} the

516: vertical axis for $\gamma$ is reversed.

517: 

518: Careful examination of the plots in Fig. \ref{fig:lightcurves} and

519: \ref{fig:glogf} reveals that

520: \begin{itemize}

521: \item Most of the emission spikes are well represented by straight lines in

522:   $\gamma$-$\log I_{50}$ during the rise and decay phase. The decay phases are

523:   sometimes flatter in $\gamma$-$\log I_{50}$ than the corresponding rise phases

524:   (e.g. panel C), but the opposite is also observed (panel A). The event shown

525:   in panel E shows some significant deviations from the piecewise straight

526:   trend.

527: \item Spectral variability is stronger at the beginning of the event (panels A,

528:   B, C, F).

529: \item In the late phase of the events a slower varying component is seen,

530:   piecewise straight in $\gamma$-$\log I_{50}$, mostly nearly flat, slowly

531:   hardening (panels B, D), slowly softening (panels A, F), staying at an

532:   approximate constant hardness (panel C), or a mixture of the above (panel E).

533: \item During the rise phase up to the strongest peak, the hardness tends to

534:   increase from peak to peak (panels A, C). Events are softer in the beginning

535:   (all panels).

536: \end{itemize}

537: 

538: Figure \ref{fig:histoslope} shows the distribution of the $\gamma$-$\log I_{50}$ slopes

539: (parameter $a$ in Eqs. \ref{eq:lintrendglogf} and \ref{eq:pivslope}) in

540: the rise and decay phases of all emission spikes in the 5

541: events. If the flux increases, a negative slope represents spectral hardening and a

542: positive slope a softening. The opposite happens during a decay phase. SHS peaks

543: have $a<0$ both in the rise and decay phase.

544: 

545: The separate histograms for rise and decay phase differ marginally. Spectral

546: hardenings during rise have a slightly steeper $\gamma$-$\log I_{50}$ slope than

547: softenings during decays (SHS peaks). Quantitatively, the average value of $a$

548: when restricted to negative values is of $-0.70 \pm 0.06$ (the uncertainty is

549: the standard error of the mean) during the rise phases and $-0.51 \pm 0.08$

550: during the decay phases (after removing an outlier with slope -3.2). The average

551: of $a$ for the combined set of rise and decays is $-0.60 \pm 0.05$,

552: corresponding to an average pivot point energy (see App. B) of $\eps_*=9.4 \pm

553: 1.3$ in agreement with \cite{grigis04}.

554: 

555: Examination of spectral hardenings during decay and softenings during rise

556: phases (trends opposite to SHS), restricted to the range 0--3, yield an average

557: $a$ value of 0.66$\pm$0.11 and 0.88$\pm$0.18, respectively. These values are not

558: significantly different from each other or from the corresponding absolute value

559: of the $a<0$ averages.

560: 

561: 

562: \section{Imaging results}

563: 

564: 

565: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

566: %

567: %Table 2: motion at the onset of the hardening

568: %

569: 

570: 

571:  

572: \begin{table*}

573: {\scriptsize %only when referee format is used --> otherwise table

574: %don't fit in page. Fits when using the emulateapj package.

575: \caption{Source motion near the onset of hardening.}

576: 

577: \label{tab:hardeffect}

578: \centering

579: %\hspace{-1.2cm}

580: \begin{tabular}{l l l l}

581: \hline\hline

582: Event & Onset of  & Footpoints  & Coronal source \\

583:       & hardening & motion      & motion         \\

584: \hline 

585: 07-NOV-2004 & 16:26 & Jump in the position of the eastern FP %1 min before onset  

586:                     & uncertain (bad images at\\

587:             &       & \ \ Change in the direction of motion of the western FP % 1 min before the onset

588:                     & \ \ low energies after onset) \\ 

589: 10-NOV-2004 & 02:12 & Jump in the position of the two brighter FPs & stationary   \\

590: 17-JAN-2005 & 09:49 & Nearly continuous motion of both FPs & continuous motion \\

591: 19-JAN-2005 & 08:27 & Continuous motion of the northern FP, slowing after the onset &

592: continuous motion upward  \\

593:             &       & Nearly stationary position of the southern FP & \\

594: 20-JAN-2005 & 06:49 & Reversal in the direction of motion of both FPs, slowing down

595: afterwards & continuous motion upward\\

596: \end{tabular}

597: }%end footnotesize

598: \end{table*}

599:  

600: %

601: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

602: 

603: 

604: 

605: Is there a connection between spectral hardening and X-ray source geometry

606: indicating a different acceleration site? The position of the hard X-ray

607: footpoint sources were investigated in CLEAN images, using detectors 3 to 8 with

608: a cadence of 60 seconds. In particular, we looked for differences in source

609: positions and velocities at the onset and during the period of general

610: hardening.

611: 

612: Figure \ref{fig:evc} shows an overview of the position of the thermal

613: and non-thermal sources during the events. The event of 10-NOV-2004 is not shown

614: because it lacks a coordinated evolution of source positions.

615: 

616: The event of 19-JAN-2005 is particularly interesting, as it was well observed

617: and the footpoints (FP) clearly move along the ribbons noticeable in a TRACE

618: image at 1600 \r{A} \citep{2008ApJ...673.1169S}. The motion is fast at the

619: beginning and later slows down. The thermal source has the form of a loop,

620: rising throughout the event.

621: 

622: Figure \ref{fig:fpmov_ccross} shows the displacement of the northern FP source

623: in the directions parallel and perpendicular to the northern ribbon\footnote{The

624:   parallel and perpendicular directions are defined relative to the direction of

625:   the regression line obtained by least-square fitting the positions of the

626:   hard X-ray footpoints independently in each ribbon.}. The FP source starts with

627: a velocity of 50 km s$^{-1}$ along the ribbon and slows down continuously during

628: the SHS phase until the onset of the hardening phase, when it becomes nearly

629: stationary. There is no evidence of an abrupt transition between the two

630: regimes. Furthermore, 19-JAN-2005 is the only event clearly showing footpoints

631: drifting apart and coming to a stop at the onset of hardening.

632: 

633: The other events show other positional changes. In the following the geometrical

634: behavior of the sources and their evolution are described shortly for all the

635: events. We distinguish between footpoint sources and coronal sources. Because of

636: projection effects, we cannot reliably reconstruct the three dimensional

637: structure of the sources. However we know from limb event observations

638: \citep{battaglia06} that FP sources are mainly non-thermal and well-observed

639: above 20-30 keV, whereas coronal sources are mainly thermal and well-observed

640: below 20-30 keV. Therefore, we assume in the following that the thermal source

641: is coronal and that non-thermal sources are footpoints.

642: 

643: 

644: {\bf 07-NOV-2004:} This events features two footpoints and a coronal source.

645: The eastern FP moves from W to E from 16:20 to 16:24 UT, jumps back to near the starting

646: position and moves again from W to E from 16:25 to 16:30. The western FP moves

647: slightly from SW to NE from 16:21 to 16:24, then changes direction with a slight 

648: jump to W, and slowly moves to NW from 16:25 to 16:30. The western FP is brighter

649: than the eastern FP before 16:24 and dimmer after 16:25.  The thermal source is

650: located farther E than the FPs and moves slightly to N from 16:19 to 16:25, and is

651: not clearly seen in the images afterwards, due to the insertion of RHESSI's

652: thick attenuators.

653: 

654: The jump in position around 16:25 roughly coincides with the time at which the

655: hardening starts, possibly indicating that another loop is actively accelerating

656: electrons, but the expansion of the loop, as suggested by the FP motions,

657: continues during the hardening phase, contrarily to what has been observed in

658: the event of 19-JAN-2005.\\

659: 

660: 

661: {\bf 10-NOV-2004:} This event has a very complicated FP morphology, with

662: sources and source-pairs appearing in many different places. It is not possible

663: to find a well-defined source motion like in the simpler cases with only two

664: footpoints. Here the sources seem to jump around as new footpoints in a

665: different position become brighter and outshine the old ones. Important shifts

666: in position occur at 02:08 and 02:12. The latter shift happens at the same time

667: as the onset of hardening.\\

668: 

669: 

670: {\bf 17-JAN-2005:} Three pairs of FPs are seen. The northern pair is stationary

671: and seen from 09:43 to 09:45. The southern pair consists of an eastern FP moving

672: to SE from 09:45 to 10:05, and a western FP moving to N from 09:43 to 09:47, then

673: shifting to W (09:55) and moving very slowly to N until 10:05. The last pair of FP

674: is to the east and stationary from 10:11 to 10:29.  Coronal sources are seen in

675: two locations: one to the N of the southern FP pair, moving to N from 09:46 to

676: 09:57, and the second to the NW of the easternmost FP pair, nearly stationary

677: from 10:16 to 10:30.

678: 

679: There is no clear signature of a discontinuity or a change of behavior

680: happening around 09:50, when the hardening starts.\\

681: 

682: 

683: {\bf 19-JAN-2005:} Two footpoints are seen with a loop-shaped coronal source

684: between them. The northern FP moves to NE from 08:12 to 08:30 (covering nearly 60

685: arcseconds) while the southern FP moves, slower, to SE. In the meantime, the

686: loop-shaped coronal source moves to NW (indicating that it is 

687: rising). After 08:30 (only 3 minutes after the onset of hardening) the northern

688: FP is much slower. The coronal source keeps moving to NNW. In the next RHESSI

689: orbit (after 09:30), the FP sources can still be seen near the old positions at

690: 08:30. The northern FP is nearly stationary from 09:33 to 09:59, while the southern

691: FP very slightly moves to W, and the coronal source slightly moves to N.\\

692: 

693: 

694: 

695: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

696: %

697: %Figure 5: Imaging of events

698: %

699:  

700: \begin{figure*}

701: %\plotone{im_c.eps}

702: \epsscale{0.55}

703: \begin{tabular}{ccc}

704: \plotone{f5_part1.eps} & \hspace{-2cm}\plotone{f5_part2.eps} & \\

705: \plotone{f5_part3.eps} & \hspace{-2cm}\plotone{f5_part4.eps} & \\

706: \end{tabular}

707: %

708: \caption{ RHESSI hard X-ray positions of the non-thermal footpoint sources

709:   (stars) and the thermal coronal source (diamond) for 4 events. The energy

710:   range for the footpoint position is 50-100 keV band (or 25-50 keV at times

711:   where the 50-100 keV emission is too faint for imaging). The energy range for

712:   the coronal source position is 12-25 keV. Systematic motions are indicated by arrows.}

713: %

714: \label{fig:evc}

715: \end{figure*}

716: 

717: {\bf 20-JAN-2005:} This near-limb event features two FPs and a loop-like coronal

718: source. The southern FP moves to W while the northern FP moves to E. The

719: eastward motion of the northern FP is not continuous: at 06:49 it reverses and

720: recedes until 06:55, when another sources appears 20$^{\prime\prime}$ to E. The

721: double structure lasts until 07:01, when the easternmost source fades away. The

722: coronal source moves to NW and rises throughout the event, slowing down towards

723: the end. The reversal coincides with the start of the hardening phase.\\

724: %confirmed

725: 

726: The observed behavior at the onset of hardening for all events is reported in

727: Table \ref{tab:hardeffect}. It summarizes the analysis of the source motions in

728: an interval of time spanning 4 minutes, centered on the onset of hardening. We

729: conclude from the imaging observations that there is no universal trend

730: holding for all events. Sometimes, there seems to be a switch to a different

731: loop system near the beginning of the hardening phase.  On the other hand, such

732: jumps can also be seen during the SHS phase of the events, so they need not be

733: significant. There is also some indication that the FP motion is slower during

734: the hardening phase, but again this does not hold for all events. In the event

735: of 19-JAN-2005, with a simple geometry and well observed, the change in spectral

736: behavior leading to the hardening phase does not have an impact on the

737: morphology of the hard X-ray sources seen by RHESSI.

738: 

739: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

740: %

741: %Figure ?: motion & cross-corr coeffs.

742: %

743:  

744: \begin{figure}

745: %\resizebox{\hsize}{!}{\includegraphics{fig4.eps}}

746: \plotone{f6.eps}

747: %

748: \caption{Movement of the northern X-ray footpoint source for the event of

749:   19-JAN-2006. The upper curves show the motion in the two components parallel

750:   (top curve) and perpendicular (middle curve, multiplied by a factor of 3) to

751:   the northern TRACE 1600 \r{A} ribbon. The bottom curve displays the cross

752:   correlation coefficient between hard X-ray spectral index and flux, showing

753:   the onset of spectral hardening around 08:27.}

754: \label{fig:fpmov_ccross}

755: \end{figure}

756:  

757: %

758: %

759: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

760: 

761: 

762: \section{Modeling the SHH phase}

763: \label{modeling}

764: %

765: 

766: \cite{grigis06} showed that the soft-hard-soft trend is expected from a

767: transit-time damping stochastic acceleration model that includes escape of

768: particles from the accelerator. The hardness is controlled by how fast the

769: particle gain energy and how long they are trapped in the accelerator. Harder

770: spectra result from longer dwelling times of the electrons in the accelerator

771: and higher acceleration efficiency. These conditions also allow a larger

772: population of high-energy electrons to build up, leading to increased hard X-ray

773: emission from the accelerator, identified as a part of the looptop source.  This

774: basic model predicts that harder spectra also have larger hard X-ray flux, but

775: cannot explain the soft-hard-harder trend seen as the flux decays, because 

776:   these observations associate harder spectra with smaller flux.

777: 

778: To fit the observed SHS behavior, \cite{grigis06} had to assume that electrons

779: are trapped below a certain threshold energy $E_\mathrm{T}$ and cannot leave the

780: accelerator. The escaping electron population has a low-energy cutoff at

781: $E_\mathrm{T}$. Then the photon spectra \emph{of the footpoints}, which dominate

782: the non-thermal emission, harden below $E_\mathrm{T}$.

783: 

784: In the following, a simple extension of the basic stochastic acceleration model

785: is presented which could lead to the observed spectral hardening. We introduce

786: the additional assumption that $E_\mathrm{T}$ increases with time in the SHH

787: phase. Therefore, the photon spectrum below $E_\mathrm{T}$ hardens while at the

788: same time the flux arriving at the footpoints decreases. The important point

789: here is that this also happens if the electron spectral index $\delta$ and the

790: electron flux normalization in the acceleration region are constant. Thus the

791: new variable does not contradict the basic properties of the stochastic

792: acceleration model. This is shown in Fig. \ref{fig:spmod}, where footpoint

793: photon spectra have been computed from a given electron distribution with

794: various $E_\mathrm{T}$. As expected, the photon flux decreases with increasing

795: $E_\mathrm{T}$. Note that the photon spectra have a nearly constant but slightly

796: increasing hardness while the flux decreases.

797: 

798: This extension is compared numerically to the late phase of the 19-JAN-2005

799: event. For simplicity, an electron distribution with constant spectral index and

800: flux normalization is assumed to escape from the acceleration region. It has a

801: low-energy cutoff $E_\mathrm{T}$ due to trapping in the

802: accelerator. $E_\mathrm{T}$ increases with time. Obviously, we do not expect

803: such a simple scenario to reproduce all the details of the spectral

804: evolution. The question is, how much of the observed features can be explained

805: with this simplest extension of the existing acceleration model.

806: 

807: 

808: 

809: A non-isotropic energy distribution of fast electrons with a positive slope in

810: energy is known to be unstable towards growing plasma waves. Therefore our

811: scenario also includes an alternative electron distribution featuring a turnover

812: at $E_{T}$, that is, a flat distribution below $E_{T}$, instead of a

813: cutoff. The electron distributions with cutoff and turnover are assumed as,

814: 

815: \begin{eqnarray}

816: \label{eqn:modelsp}

817:   F_\mathrm{CUTOFF}(E) & = & \left\{

818:        \begin{array}{lll}

819:          F_{E_0}\left(\displaystyle\frac{E}{E_0}\right)^{\delta} 

820:                               & \mathrm{if} & E\ge E_\mathrm{T}\\

821:          0 &\mathrm{if} & E< E_\mathrm{T} \\

822:          \end{array}\right.\\

823:   F_\mathrm{TURNOVER}(E) & = &  \left\{ \begin{array}{lll}        

824:     F_{E_0}\left(\displaystyle\frac{E}{E_0}\right)^{\delta} 

825:     & \mathrm{if} & E\ge E_\mathrm{T}\\

826:     F_{E_0}\left(\displaystyle\frac{E_{T}}{E_0}\right)^{\delta} &\mathrm{if} & E< E_\mathrm{T}

827:          \end{array}\right.

828: \end{eqnarray}

829: 

830: The free parameters are the electron spectral index $\delta$, the electron flux

831: normalization $F_{E_0}$ (electrons s$^{-1}$ keV$^{-1}$), the cutoff or turnover

832: energy $E_\mathrm{T}$. The reference energy $E_0$ is fixed at 50~keV.

833: 

834: 

835: An exhaustive description of the method used for the comparison of this

836: simple model with the observation and finding the best fit is given in

837: App. \ref{app:b}.

838: 

839: 

840: The late phase of the event of 19-JAN-2005, from 09:32 to 10:02, comprises 30

841: minutes of continuous hardening, and therefore is well suited for the comparison

842: with the variable cutoff model. Figure \ref{fig:vdc} shows the comparison. The

843: observed values of $\gamma$ and $\eta$ are plotted as a function of $I_{50}$

844: together with the best-fit model curve for both the cutoff and turnover electron

845: spectra. For the cutoff model, the best fit values of the parameters are:

846: spectral index $\delta= 6.52$, flux normalization at 50 keV $F_{50}=4.50\cdot

847: 10^{33}$ electrons s$^{-1}$ keV$^{-1}$. In the course of the decay,

848: $E_\mathrm{T}$ increases from 98 keV to 159 keV (thus yielding a photon spectral

849: index $\gamma$ in the range between -2.5 and -1.9). The corresponding values for

850: the turnover model are: $\delta=6.11$, $F_{50}=3.67\cdot 10^{33}$ electrons

851: s$^{-1}$ keV$^{-1}$, while $E_\mathrm{T}$ increases from 127 to 224 keV.

852: 

853: The photon spectra from both the cutoff and turnover model are curved downward

854: in the fitted energy range. Thus the spectral curvature, $\eta$, is

855: negative. It is observed to be between 0 and -0.3, whereas the model spectra

856: have values in the range from -0.55 to -0.25 (Fig. \ref{fig:vdc}, bottom)

857: and thus are significantly more curved.

858: 

859: In the cutoff distribution, the total fluxes of electrons escaping the

860: acceleration site are initially $9.9\cdot10^{32}$ electrons s$^{-1}$ above

861: $E_\mathrm{T}=98$ keV. This reduces to $6.6\cdot10^{31}$ electrons s$^{-1}$ for

862: $E_\mathrm{T}=159$ keV in the course of the decay. The total injected power

863: reduces from $1.9\cdot 10^{26}$ erg s$^{-1}$ to $2.1\cdot 10^{25}$ erg

864: s$^{-1}$. In the turnover model, the total number of particles diminishes from

865: $2.3\cdot 10^{33}$ to $1.3\cdot 10^{32}$, and the injected power from $1.9\cdot

866: 10^{26}$ erg to $2.8\cdot 10^{25}$ erg.

867: 

868: 

869: 

870: \section{Discussion}

871: \label{discussion}

872: 

873: 

874: Spectroscopic RHESSI observations are well suited to study the spectral

875: evolution during the main phase of large flares.  The path observed in the

876: $\gamma$ vs. $\log I_{50}$ plots for the events is not simple. However, it can

877: be broken down reasonably into a superposition of linear sections during flux

878: rise and decay phases. While not all rise or decay phases can be so decomposed,

879: this simple description is adequate for most of them, and permits comparison of

880: observations and theory.

881: 

882: There is a difference between the results reported here and the results from

883: \cite{grigis04} in the asymmetry between rise and decay phases in SHS peaks. The

884: previous results indicated that decay phases are steeper in the $\gamma$

885: vs. $\log I_{50}$ plot than rise phases. We find the

886: opposite. The reason is probably the selection bias: here we selected 

887: specifically events showing hardening. This hardening trend sometimes

888: overlays SHS peaks, giving rise to a soft-hard-less-soft pattern.

889: 

890: The hard X-ray images during the events show the usual morphology of hard X-ray

891: solar flares: a low-energy coronal source and two or more high-energy

892: footpoint sources. The position of the footpoint sources is strongly variable:

893:  it either moves smoothly or jumps from location to location. This reflects

894: changes in the connection between the accelerator and the chromosphere, as well

895: as in the location of the accelerator itself.

896: 

897: The behavior observed in the images cannot be reduced to one simple scenario

898: valid for all events. However, the observations suggest that there is

899: no clear separation between the SHS and the hardening phases: the former seems

900: to smoothly merge into the latter. Even in the cases where the emission jumps at

901: the onset of hardening (Table \ref{tab:hardeffect}), the footpoint behavior

902: seems not to change radically.

903: 

904: 

905: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

906: %

907: %Figure 7: modeling

908: %

909:  

910: \begin{figure}

911: %\resizebox{\hsize}{!}{\includegraphics{figspmod.ps}}

912: \plotone{f7.eps}

913: %

914: \caption{ 

915:   Photon spectra by thick target emission (at footpoints) from a

916:   constant power-law electron distribution having spectral index of

917:   $\delta=-6.5$ in the acceleration region. The low-energy cutoff,

918:   $E_\mathrm{T}$, increases from the top to the bottom curve from 0, 100,

919:   120, 140, to 160 keV. 

920: %

921: }

922: \label{fig:spmod}

923: \end{figure}

924:  

925: %

926: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

927: 

928: 

929: Alternatives to the scenario presented in Sect. \ref{modeling} are

930: conceivable. In particular, electron storage in the corona and slow release

931: during the decay could be a possibility. As Coulomb interactions are faster at

932: low particle energies, the spectrum would harden with time while the released

933: flux decreases.  Noting that the hardening phase in the 19-JAN-2005 event lasts

934: more than 30 minutes and that the decay of the flux in time is nearly exponential

935: (as seen by the fact that the $I_{50}$ line in Fig. \ref{fig:lightcurves}, panel

936: D, is nearly straight), the total number of injected electrons can be computed

937: from the total electron fluxes at the start $F_\mathrm{BEG}$ and at the end

938: $F_\mathrm{END}$.

939: \begin{equation}

940:   F_\mathrm{TOT}=\Delta t

941:       \frac{F_\mathrm{END}-F_\mathrm{BEG}}{\log\displaystyle\frac{F_\mathrm{END}}{F_\mathrm{BEG}}}\,,

942: \end{equation}

943: where $\Delta t$ is the observed duration (here 30 minutes) and $F_\mathrm{TOT}$

944: the total injected flux. From the observed values, we get $F_\mathrm{TOT}\simeq

945: 1\cdot 10^{36}$ electrons for the turnover model and $F_\mathrm{TOT}\simeq

946: 6\cdot 10^{35}$ electrons for the cutoff model. These numbers do not seem

947: extraordinarily high, but it should be noted that all these electron have

948: energies above 100 keV.

949: 

950: Would such a large population of electrons be seen as a coronal hard X-ray

951: source in the 50 - 100 keV band? The luminosity depends on the volume and

952: density in the storage region. The observed footpoints in the hardening phase

953: are separated from each other by approximately 60$^{\prime\prime}$, indicating a

954: medium sized loop. Therefore, it would be visible on RHESSI images unless it

955: were excessively under-dense. Thus we reject the storage model.

956: 

957: 

958: Both the cutoff and turnover model are able to reproduce the observed $\gamma$

959: vs. $\log I_{50}$ trend, but fail to reproduce the correct spectral curvature

960: $\eta$ (Fig. \ref{fig:vdc}). Although the observation of $\eta$ is more difficult in the decay phase

961: due to the lower signal-to-background ratio, the difference between the cutoff

962: and turnover models and the observed points is significant. The value of the

963: parameter $\eta$ depends on the energy interval chosen for the fitting of the

964: model photon spectra (20-80 keV in our case). A lower maximum energy of this

965: interval produces lower model values for $\eta$.

966: 

967: We note furthermore that if the accelerator is inhomogeneous, the electron

968: spectrum at the footpoints is the superposition of different components with

969: different values of the low-energy cutoff or turnover $E_\mathrm{T}$. The

970: superposition of spectra that curve at different energies is less curved than

971: individual spectra. Figure \ref{fig:spmod} suggests that the superposition of

972: components with different spectral shape may in fact reduce of the total

973: spectral curvature. Thus we consider the disagreement in curvature not 

974: crucial to reject the model.

975: 

976: 

977: 

978: %

979: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

980: %

981: 

982: \section{Conclusions}

983: \label{conclusion}

984: 

985: We have presented results of spectroscopy, imaging and simple modeling of

986: spectral hardening observed occasionally in the hard X-ray emission of large

987: flares. The main conclusions are:

988: 

989: 

990: \begin{itemize}

991: \item The flares selected for the presence of a hardening phase also show

992:   soft-hard-soft behavior, at least initially. The hardening starts at or

993:   after the largest peak of the flares. In 3 out of 5 events it starts 2 to 6

994:   minutes after the onset of a CME.

995: \item Similar to SHS peaks, hardening phases can usually be described by

996:   piecewise linear sections in a plot of spectral index vs. logarithmic flux.

997: \item There is no clear trend relating the behavior of hard X-ray footpoint sources with

998:   the spectral evolution that would be valid for all events. Sometimes the location of the

999:   emission shifts when the hardening starts, in other events it does not.

1000: \item In the event of 19-JAN-2005, there are only two well-defined

1001:   footpoint sources throughout the whole event. No

1002:   discontinuity is observed in the motion at the onset of hardening, but a general trend

1003:   of slowing down, such that the FPs become nearly stationary during the decay phase.

1004: \item In 3 out of 5 flares, the coronal source moved continuously during

1005:   the onset of the hardening. This motion was directed upwards in

1006:   two near-limb events.

1007: \end{itemize}

1008: 

1009: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1010: %

1011: %Figure 8: modeling

1012: %

1013:  

1014: \begin{figure}

1015: %\resizebox{\hsize}{!}{\includegraphics{figmodel.ps}}

1016: \plotone{f8.eps}

1017: %

1018: \caption{ 

1019:   Observed values of $\gamma$ vs. $I_{50}$ (\emph{top panel}) and $\eta$

1020:   vs. $I_{50}$ (\emph{bottom panel}), represented by asterisks. The 

1021:     best-fit model curves (continuous for the cutoff model, dashed for the

1022:   turnover model) are expected from model with constant electron hardness and

1023:   flux, but rising low energy cutoff or turnover energy $E_\mathrm{T}$.}

1024: \label{fig:vdc}

1025: \end{figure}

1026:  

1027: %

1028: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1029: 

1030: 

1031: 

1032: 

1033: In the sample studied, we find a surprising lack of detailed correlation between

1034: spectral and spatial behavior. It is similar to what has been observed by

1035: \cite{2005ApJ...625L.143G} in a smaller flare (M6) featuring strong

1036: footpoint motions and hardening at the end.

1037: 

1038: The main question addressed in this paper is whether the SHS peaks and the

1039: hardening phases are the results of two different acceleration mechanisms. The

1040: results support the view that the same acceleration mechanism changes gradually

1041: in the later phase of the flare. This change has clear effects on the spectrum,

1042: but a more indirect and subtle influence on the source position. The operation

1043: of a second acceleration process later in the flare cannot be ruled out,

1044: however. Nevertheless, we have found strong evidence that there is a gradual

1045: change in the accelerator, transforming its behavior from impulsive (showing up

1046: as SHS peaks) to gradual (hardening phases). This is substantiated by the

1047: observations of the superimposition of SHS peaks with a continuous hardening

1048: trend and of the smooth footpoint motions during the onset of hardening.

1049: 

1050: 

1051: The reason for the association with interplanetary proton events

1052: \citep{kiplinger95} remains to be explored. As an aside we may note that when

1053: the footpoints drift apart, acceleration takes place in larger and larger

1054: loops. In a stochastic acceleration framework, the acceleration efficiency of

1055: electrons in larger loops is reduced, while ions can be more efficiently

1056: accelerated \citep{emslie04}. Since hardening trends are well correlated with

1057: the occurrence of interplanetary energetic protons events, it is possible that

1058: the very conditions that are responsible for the hardening trends favor

1059: acceleration of protons, which may then escape into interplanetary space, with

1060: the CME controlling their release rather than acceleration

1061: \citep{2006A&A...445..715S}.

1062: 

1063: 

1064: The observed motion of footpoints suggests that different coronal loops may be

1065: involved in particle acceleration during a flare. They will have different

1066: physical properties such as size, density, and magnetic field. The overall

1067: magnetic geometry of the active region will determine which loops reconnect at

1068: which time, sometimes giving rise to an orderly motion of footpoints, sometimes

1069: generating a more chaotic situation. The data suggest that as the reconnection

1070: process proceeds, some physical parameters of the acceleration site changes in

1071: such a way as to favor the production of harder spectra, rather than having a

1072: totally new process (say, shock acceleration) taking over in the decay phase.

1073: 

1074: 

1075: 

1076: 

1077: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1078: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1079: 

1080: \acknowledgements

1081:   We thank S. Krucker, Y. Su and the participants of the 7th general RHESSI

1082:   Workshop for useful discussions.  The analysis of RHESSI data at ETH Zurich is

1083:   partially supported by the Swiss National Science Foundation (grant

1084:   nr. 20-67995.02). This research has made use of NASA's Astrophysics Data

1085:   System Bibliographic Services.  The SOHO LASCO CME catalog is generated and

1086:   maintained at the CDAW Data Center by NASA and The Catholic University of

1087:   America in cooperation with the Naval Research Laboratory.

1088:   {\it Facilities:} \facility{RHESSI}

1089: 

1090: 

1091: 

1092: \appendix

1093: \section{Hard X-ray model fitting}

1094: \label{app:a}

1095: 

1096: The hard X-ray spectrum observed in solar flares consists of two distinct

1097: components at low energies (that is, below 10-40 keV) and high energies. The

1098: properties of the low-energy component are typical of thermal emission of a hot

1099: plasma with temperature of 10 to 40 MK. In particular, the spectrum steepens

1100: with energy, and the temporal evolution follows the (cooler) thermal plasma

1101: observed in soft X-rays and EUV. 

1102: 

1103: The high-energy component behaves differently and is hence dubbed

1104: non-thermal. It is harder then the thermal component and is usually fitted with

1105: a power-law function of the energy with 2 free parameters. However, sometimes

1106: the observed spectrum steepens at higher energies. In the literature, this is

1107: usually accounted for by using a broken power-law model

1108: \citep[e.g.][]{dulk92,battaglia05}. There are some disadvantages in the broken

1109: power-law model: a) it is not physical, in the sense that any continuous

1110: electron distribution emitting X-rays should generate a differentiable photon

1111: spectrum, and b) the location of the break-point is poorly determined by the

1112: observations.

1113: 

1114: 

1115: We argue that there is a simpler extension to the power-law model which both

1116: turns down at higher energies and is smooth. Recalling that a power-law function

1117: plotted in log-log space is a straight line, we choose as a ``natural''

1118: extension to the next order a parabolic model in log-log space, described by the

1119: function

1120: %

1121: \begin{equation}

1122:   \label{eq:spmodel}

1123:   I(\eps)=I_{\eps_0}\cdot\left(\frac{\eps}{\eps_0}\right)^{\displaystyle \gamma -\eta\log\left(\eps/\eps_0\right)}\,.

1124: \end{equation}

1125: %

1126: The 3 model parameters are the normalization $I_{\eps_0}$, amounting to the flux

1127: at the (fixed) normalization energy $\eps_0$, the spectral index $\gamma<0$ and

1128: the parabolic coefficient $\eta$, which we will refer to as the \emph{spectral

1129:   curvature} although, strictly speaking, the geometric curvature of the

1130: parabola is not constant, but equals $-2\eta$ in the vertex and vanishes at

1131: infinity.

1132: 

1133: 

1134: In the special case $\eta=0$, the (unbroken) power-law model is recovered. We

1135: note here that a log-parabolic model has been used previously to describe

1136: observed X-ray spectra of pulsars \citep{massaro00}.

1137: 

1138: In summary, the reasons for preferring the log-parabolic model over the more

1139: usual broken power-law are:

1140: \begin{enumerate}

1141: \item It is simpler than the broken power-law, as it allows only 3 instead of 4

1142:   free parameters.

1143: \item For the vast majority of the time intervals, it produces similar values of

1144:   $\chi^2$ as the broken-power law.

1145: \item It is differentiable, therefore there exists a continuous electron

1146:   spectrum producing the photon spectrum. This is not the case for the broken

1147:   power-law, where a discontinuity is needed in the electron spectrum, which would 

1148:   quickly be eliminated by kinetic plasma processes. The spectral index

1149:   increases linearly with $\log\eps$.

1150: \item It allows a better comparison with acceleration models which naturally

1151:   produce slightly curved electron spectra (like stochastic acceleration).

1152: \end{enumerate}

1153: 

1154: Therefore, we fit the spectra to a photon model with an isothermal component at

1155: lower energies and a log-parabolic component as given above at higher energies.

1156: The background is taken into account in the following way: the pre-event and

1157: post-event background spectra are measured and averaged (in some cases, particle

1158: contamination prevented to obtain both of them, and only one was taken instead),

1159: yielding a reference background spectrum. The reference background spectrum is

1160: multiplied with a free parameter $\lambda$ and added to the model

1161: spectrum. $\lambda$ is fitted together with the other model parameters. 

1162: 

1163: In large flares, sometimes an additional hard and weak $\gamma$-ray emission

1164: from electrons is observed above 200 keV \citep{2008ApJ...678L..63K}. In this

1165: paper we do not report on the properties of this emission since it is not

1166: related to the hardening of the spectrum at lower energies. Does this component

1167: negatively affects our fittings?  Examination of the fitted spectra reveal that

1168: the fittings account for this weak, hard emission by an increase of the

1169: parameter describing the strength of the background ($\lambda$) by a factor of

1170: about 1 to 2. The fittings are of good quality because the background is also

1171: weak and hard. This erroneous increase in the strength of the background has no

1172: effect at lower energies, since there the hard X-ray flux is stronger by orders

1173: of magnitude.

1174: 

1175: \section{Pivot point and parabolic fitting}

1176: \label{app:b1}

1177: 

1178: 

1179: A linear dependence of $\gamma$ vs. $\log I_{\eps_0}$ with negative slope can be

1180: interpreted geometrically as a fixed intersection point of the various power-law

1181: spectra at different times. The intersection in the spectral plane ($I$

1182: vs. $\eps$) is called the \emph{pivot point}, located below the reference energy

1183: $\eps_0$ \citep{grigis05}. Similarly, a linear dependence with positive slope

1184: can be interpreted by a pivot point at energy larger than $\eps_0$. If fitting a

1185: log-parabolic spectrum, it is the tangents to the spectrum at $\eps_0$ in

1186: log-log space that are intersecting, rather than the curves themselves.

1187: 

1188: The description in terms of a pivot point has the advantage that it does not

1189: depend on the choice of the reference energy $\eps_0$. On the other hand, the

1190: pivot point energy jumps from $0$ to $+\infty$ when the slope in $\gamma$-$\log

1191: I_{\eps_0}$ goes from negative to positive.  The relation between the pivot

1192: point coordinates $\eps_*,I_*$ and the line parameters $a,b$ in

1193: Eq. \ref{eq:lintrendglogf} are given by

1194: \begin{equation}

1195: \label{eq:pivslope}

1196: \qquad a=\frac{1}{\log \left(\eps_*/\eps_0\right)}\qquad

1197: \qquad b=\frac{-\log{I_*}}{\log \left(\eps_*/\eps_0\right)}\,.

1198: \end{equation}

1199: 

1200: Since the spectra are curved in log-log space, the local spectral

1201: index~$\gamma^\prime$, defined as the logarithmic derivative of the spectrum

1202: %

1203: \begin{equation}

1204: \gamma^\prime(\eps)=\diff{\log I(\eps)}{\log \eps}=\gamma - 2\eta\log(\eps/\eps_0)\,,

1205: \end{equation}

1206: %

1207: is energy dependent. The spectral parameter $\gamma$ is equal to the local

1208: spectral index $\gamma^\prime$ at the reference energy $\eps_0=50$~keV. The

1209: presence of a strong correlation of the time series of $\gamma$ and $\log

1210: I_{50}$ does not necessarily imply a strong correlation of the time series of

1211: $\gamma^\prime(\eps)$ and $\log I(\eps)$ at energies $\eps\ne \eps_0$.  In fact,

1212: if the spectra have a common pivot point at $\eps_*<\eps_0$, the correlation of

1213: $\gamma^\prime(\eps)$ and $\log I(\eps)$ weakens near the pivot point, and turns

1214: into an anticorrelation for $\eps<\eps_*$. On the other hand, if a pivot point

1215: exist at $\eps_*>\eps_0$, then the anticorrelation between $\gamma^\prime(\eps)$

1216: and $\log I(\eps)$ for $\eps\simeq\eps_0<\eps_*$ transforms itself into a

1217: correlation  at $\eps > \eps_*$.

1218: 

1219: An examination of the data for the events studied here shows that in the time

1220: intervals when the correlation coefficient between $\log I_{50}$ and $\gamma$ is

1221: near 1, the correlation coefficient between $\gamma^\prime(\eps)$ and $\log

1222: I(\eps)$ is approximately constant at energies higher than 30-40 keV However, it 

1223: shows a strong decrease to values close to -1 at lower energies such that the

1224: transition takes place around 5-20 keV, near the pivot point position.  On the

1225: other hand, when the correlation coefficient is near -1 at 50 keV we observe

1226: that the correlation coefficients rises toward 1 at higher energies, as these

1227: events tend to have a pivot point at energies larger than 50 keV.

1228: 

1229: Therefore, the energy dependence of the correlation in the data follows a

1230: similar pattern as the one expected for non-curved spectra. This is

1231: because the observed spectra are not strongly curved: $\eta<0.25\gamma$ in

1232: all the fitted spectra, and $\eta<0.15\gamma$ in 85\% of all fitted spectra.

1233: 

1234: 

1235: 

1236: \section{Comparison between model and observation}

1237: \label{app:b}

1238: 

1239: Comparison of non-thermal hard X-ray spectral observations and theoretical

1240: models can be performed in different ways. In our case, we have a really simple

1241: model and a time-dependent situation. The goal of the comparison is not the

1242: perfect reproduction of every single observed spectrum, but rather a coherent

1243: description of the time evolution which should be compatible with the observed

1244: data. This is implemented by the additional constraint in the model (as given by

1245: Eq. \ref{eqn:modelsp}) of keeping a constant electron spectral index $\delta$

1246: above the threshold energy $E_\mathrm{T}$. So our task consists of two steps:

1247: first, for a given value of $\delta$, we have to find the set of

1248: $E_\mathrm{T}(t)$ and $F_{E_0}(t)$ that best reproduces the data and second, we

1249: have to choose the best value of $\delta$ (for instance by running the first

1250: step for many different values of $\delta$ and pick the best one).

1251: 

1252: For the comparison of the model with the data, we generate model photon spectra

1253: emitted by the model electron spectra by computing the thick-target

1254: Bremsstrahlung emission assuming collisional energy losses and using the full

1255: relativistic Bethe-Heitler cross section \citep{BH1934} with the \cite{elwert39}

1256: correction factor.

1257: 

1258: For the first step, the obvious strategy involves forward fitting the electron

1259: model to the observed data. However, this is time consuming because it needs to

1260: be repeated for all the spectra and many different values of $\delta$. On the

1261: other hand, we have at our disposal the photon fitting parameters $I_{\eps_0}$,

1262: $\gamma$ and $\eta$ (Eq. \ref{eq:spmodel}), which are good descriptions of the

1263: observed photon spectrum as long as the reduced $\chi^2$ is around one. In this

1264: case, there is no need of additional fitting in count space: we just fit exactly

1265: the same log-parabolic model to the model photon spectra (in the energy range

1266: where the non-thermal component is seen above the thermal component and the

1267: background in the observations). This is faster, and delivers photon model

1268: parameters $I^{\mathrm{MOD}}_{\eps_0}$, $\gamma^{\mathrm{MOD}}$ and

1269: $\eta^{\mathrm{MOD}}$ as a function of the electron model parameter $F_{E_0}$,

1270: $\delta$ and $E_\mathrm{T}$.  The comparison can then be performed in the

1271: $\gamma$ vs. $I_{\eps_0}$ plot by a least square argument in two steps.

1272: 

1273: In the first step we held $\delta$ fixed and increase $E_\mathrm{T}$ to

1274:   generate a curve in the $\gamma$ vs. $I_{\eps_0}$ plot for each value of

1275: $F_{E_0}$. The normalization $F_{E_0}$ is then chosen such that it minimizes the

1276: square differences between $I_{\eps_0}$ and $I^{\mathrm{MOD}}_{\eps_0}$. The

1277: results from running step one repeatedly are a set of paths (one for each

1278: different value of $\delta$) in the $\gamma$ vs. $I_{\eps_0}$ plot, where

1279: the variable $E_\mathrm{T}$ changes along the path.

1280: 

1281: In the second step, we just select among all paths in the $\gamma$

1282: vs. $I_{\eps_0}$ space the one with the least square distances from all the

1283: observed points. This yields the best $\delta$ and $F_{E_0}$ and a range of

1284: variation of $E_\mathrm{T}$.

1285: 

1286: 

1287: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1288: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1289: %

1290: \begin{thebibliography}{}

1291: %

1292: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1293: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1294: 

1295:  

1296: \bibitem[Bai(1986)]{bai1986} Bai, T.\ 1986, \apj, 308, 912 

1297: 

1298: \bibitem[Bai \& Ramaty(1979)]{bai1979} Bai, T., \& Ramaty, R.\ 1979, \apj, 227, 1072

1299: 

1300: \bibitem[Battaglia \& Benz (2007)]{2007A&A...466..713B}%FP vs LT comparison, ITTT model 

1301:  Battaglia, M., \& Benz, A.~O.\ 2007, \aap, 466, 713

1302: 

1303: \bibitem[Battaglia \& Benz (2006)]{battaglia06}% imaging spectroscopy, shs for LT & FP sources

1304:  Battaglia, M \& Benz, A.~O. 2006, \aap, 456, 751

1305: 

1306: \bibitem[Battaglia \etal (2005)]{battaglia05}% flares of different sizes

1307:  Battaglia, M., Grigis, P.~C., \& Benz, A.~O. 2005, \aap, 439, 737 

1308: 

1309: \bibitem[Bethe \& Heitler (1934)]{BH1934}%qm computed cross section for elmg scattering

1310:  Bethe, H., \& Heitler, W. 1934, Royal Society of London Proceedings Series~A,

1311:  146, 83

1312: 

1313: \bibitem[Cliver \etal (1986)]{cliver86}% gradual events SMM

1314:  Cliver, E.~W., Dennis, B.~R., Kiplinger, A.~L., et al. 1986, \apj, 305, 920 

1315: 

1316: \bibitem[Dulk \etal (1992)]{dulk92}% Characteristics of hard X-ray spectra of impulsive solar flares

1317:  Dulk, G.~A., Kiplinger, A.~L., \& Winglee, R.~M.\ 1992, \apj, 389, 756

1318: 

1319: \bibitem[Frost \& Dennis (1971)]{frost71}% two stage acc. & SHH

1320:  Frost, K.~J., \& Dennis, B.~R.\ 1971, \apj, 165, 655

1321: 

1322: \bibitem[Emslie \etal (2004)]{emslie04}% explanation different sizes

1323:  Emslie, A.~G., Miller, J.~A., \& Brown, J.~C.\ 2004, \apjl, 602, L69

1324: 

1325: \bibitem[Elwert (1939)]{elwert39}%Elwert correction factor

1326:  Elwert, G. 1939, Ann. Phys. 34, 178

1327: 

1328: \bibitem[Grigis \& Benz (2004)]{grigis04}% SHS paper I

1329:  Grigis, P.~C. \& Benz, A.~O. 2004, A\&A, 426, 1093

1330: 

1331: \bibitem[Grigis \& Benz (2005a)]{grigis05}% SHS paper II

1332:  Grigis, P.~C., \& Benz, A.~O.\ 2005a, \aap, 434, 1173 

1333: 

1334: \bibitem[Grigis \& Benz (2005b)]{2005ApJ...625L.143G}% FP motion 

1335:  Grigis, P.~C., \& Benz, A.~O.\ 2005b, \apjl, 625, L143

1336: 

1337: \bibitem[Grigis \& Benz (2006)]{grigis06}% SHS paper III: stochastic

1338:  Grigis, P.~C., \& Benz, A.~O.\ 2006, \aap, 458, 641

1339: 

1340: \bibitem[Hudson \etal (1982)]{hudson82}% second stage acceleration in occulted flares

1341:  Hudson, H.~S., Lin, R.~P., \& Stewart, R.~T.\ 1982, \solphys, 75, 245

1342: 

1343: \bibitem[Kahler (1984)]{kahler84}% Gradual hard X-ray events and second phase particle acceleration

1344:  Kahler, S.~W.\ 1984, \solphys, 90, 133

1345: 

1346: \bibitem[Kiplinger (1995)]{kiplinger95}%spectral evolution better!

1347:  Kiplinger, A.~L.\ 1995, \apj, 453, 973 

1348: 

1349: \bibitem[Krucker et al.(2008)]{2008ApJ...678L..63K}%imaging of coronal sources >200 keV

1350:  Krucker, S., Hurford, G.~J., MacKinnon, A.~L., Shih, A.~Y., \& Lin, R.~P.\ 2008, \apjl, 678, L63 

1351: 

1352: 

1353: \bibitem[Lin \etal (2002)]{lin02}% RHESSI overview

1354:  Lin, R.~P., Dennis, B.~R., Hurford, G.~J., et al. 2002, Sol. Phys., 210, 3

1355: 

1356: \bibitem[Massaro \etal (2000)]{massaro00}%Crab pulsar

1357:  Massaro, E., Cusumano, G., Litterio, M., \& Mineo, T.\ 2000, \aap, 361, 695 

1358: 

1359: \bibitem[Ohki \etal (1983)]{ohki83}%General aspects of hard X-ray flares observed by HINOTORI Gradual burst and impulsive burst

1360:  Ohki, K., Takakura, T., Tsuneta, S., \& Nitta, N.\ 1983, \solphys, 86, 301

1361: 

1362: \bibitem[Parks \& Winckler (1969)]{parks69}%16 sec per mod xray

1363:  Parks, G.~K., \& Winckler, J.~R.\ 1969, \apjl, 155, L117

1364: 

1365: \bibitem[Saldanha \etal (2008)]{2008ApJ...673.1169S}%jan 2006 events & SEP-hardening relation

1366:  Saldanha, R., Krucker, S., \& Lin, R.~P.\ 2008, \apj, 673, 1169 

1367: 

1368: \bibitem[Piana et al.(2003)]{2003ApJ...595L.127P}%Regularized Electron Flux Spectra in the 2002 July 23 Solar Flare

1369:   Piana, M., Massone, A.~M., Kontar, E.~P., Emslie, A.~G., Brown, J.~C., \&

1370:   Schwartz, R.~A.\ 2003, \apjl, 595, L127

1371: 

1372: \bibitem[Simnett (2006)]{2006A&A...445..715S}% protons on jan 20 

1373:  Simnett, G.~M.\ 2006, \aap, 445, 715

1374: 

1375: \bibitem[Takakura \etal (1984)]{takakura84}% 1 hinotori long duration event

1376:  Takakura, T., Sakurai, 

1377:  T., Ohki, K., Wang, J.~L., Zhao, R.~Y., Xuan, J.~Y., \& Li, S.~C.\ 1984, 

1378: \solphys, 94, 359 

1379: 

1380: \bibitem[Wild \etal (1963)]{wild63}%review solar (radio mainly) bursts

1381:  Wild, J.~P., Smerd, S.~F., \& Weiss, A.~A.\ 1963, \araa, 1, 291

1382: 

1383: \bibitem[Yashiro \etal (2004)]{2004JGRA..10907105Y}% cme catalog 

1384: Yashiro, S., Gopalswamy, N., Michalek, G., St.~Cyr, O.~C., Plunkett, S.~P.,

1385: Rich, N.~B., \& Howard, R.~A.\ 2004, Journal of Geophysical Research (Space

1386: Physics), 109, 7105

1387: 

1388: 

1389: \end{thebibliography}

1390: 

1391: 

1392: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1393: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1394: %

1395: \end{document}

1396: %

1397: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1398: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1399: