astro-ph0105110/ms.tex
1: \documentclass[12pt,preprint]{aastex}
2: 
3: \begin{document}
4: 
5: \title{Temporal and Spectral Characteristics of Short Bursts
6:       from the Soft Gamma Repeaters 1806-20 and 1900+14}
7: 
8: \author{Ersin {G\"o\u{g}\"u\c{s}}\altaffilmark{1,2}, 
9: Chryssa Kouveliotou\altaffilmark{2,3},
10: Peter M. Woods\altaffilmark{2,3}, 
11: Christopher Thompson\altaffilmark{4},
12: Robert C. Duncan\altaffilmark{5},
13: Michael S. Briggs\altaffilmark{1,2}}
14: 
15: \altaffiltext{1}{Department of Physics, University of Alabama in Huntsville,
16:        Huntsville, AL 35899} 
17: \altaffiltext{2}{NASA Marshall Space Flight Center, NSSTC SD-50, 
18:        320 Sparkman Dr., Huntsville, AL 35805}
19: \altaffiltext{3}{Universities Space Research Association}
20: \altaffiltext{4}{Canadian Institute for Theoretical Astrophysics,
21:        60 St. George St., Toronto, ON, Canada M5S 3H8}
22: \altaffiltext{5}{Department of Astronomy, University of Texas, RLM 15.308, 
23:        Austin, TX 78712-1083}
24: 
25: \authoremail{Ersin.Gogus@msfc.nasa.gov}
26: 
27: 
28: \begin{abstract}
29: 
30: We study the temporal and coarse spectral properties of 268 bursts from 
31: SGR 1806$-$20 and 679
32: bursts from SGR 1900+14, all observed with the Rossi X-Ray Timing
33: Explorer/Proportional Counter Array. Hardness ratios and temporal parameters, 
34: such as T$_{90}$
35: durations and $\tau$$_{90}$ emission times are determined for these bursts.
36: We find a lognormal distribution of burst durations, 
37: ranging over more than two orders of magnitude:
38: T$_{90}$ $\sim$ 10$^{-2}$ to $\gtrsim$ 1 s, with a peak at $\sim$ 0.1 s.
39: The burst light curves tend to be asymmetrical, with more than half of
40: all events showing rise times t$_{\rm r}$ $<$ 0.3 T$_{90}$.
41: We find that there exists a correlation between the duration
42: and fluence of bursts from both sources. We also find a
43: significant anti-correlation between hardness ratio and fluence for SGR
44: 1806$-$20 bursts and a marginal anti-correlation for SGR 1900+14 events. 
45: Finally, we discuss possible physical implications of these results 
46: within the framework of the magnetar model.
47: 
48: \end{abstract}
49: 
50: \keywords{X-rays: bursts -- gamma rays: bursts -- 
51: stars: individual (SGR 1806$-$20) -- stars: individual (SGR 1900+14) }
52: 
53: 
54: \section{Introduction}
55: 
56: Soft gamma repeaters (SGRs) are a small class of objects that are
57: characterized by brief and very intense
58: bursts of soft gamma-rays and hard X-rays. 
59: They are distinguished from classical gamma-ray bursts (GRBs)
60: by their repeated periods intense activity, during which 
61: dozens of bursts with energies approaching 10$^{41}$ ergs are recorded.
62: SGR bursts have significantly softer
63: spectra than classical GRBs; the former are being well fit by an 
64: optically-thin thermal bremsstrahlung model with 
65: temperatures $kT = 20-40$ keV.  Two SGRs (0526-66 and 1900+14)
66: have emitted one giant flare each: events that are much more energetic
67: ($E \sim 10^{44-45}$ erg) and contain a very hard spectral 
68: component within the first $\sim$ 1 s (Mazets et al. 1979; Cline et al. 1981;
69: Hurley et al. 1999; Mazets et al. 1999; Feroci et al. 1999).
70: For a review of the burst and 
71: persistent emission properties of SGRs, see Hurley (2000).
72: 
73: In 1992, it was suggested that SGRs are strongly magnetized 
74: ($B \gtrsim 10^{14}$ G) neutron stars, or
75: magnetars (Duncan \& Thompson 1992, see also Pac{\`z}ynski 1992).
76: This model suggests crustquakes as a plausible trigger for the short
77: SGR bursts (as well as the giant flares):  a sudden fracture
78: of the rigid neutron star crust, driven by the build-up of 
79: crustal stresses as the strong magnetic field gradually 
80: diffuses through the dense stellar matter
81: (Thompson \& Duncan 1995 [hereafter TD95]; 1996).  The motion of the crust 
82: shears and twists the external magnetic field, and in the process
83: releases both elastic and magnetic energy.  
84: 
85: Cheng et al. (1996) studied a set of 111 SGR 1806$-$20 bursts
86: (detected with the International Cometary Explorer, ICE) and determined that
87: some of their properties, such as size and cumulative waiting-time distributions
88: are similar to those of earthquakes.
89: Recently, G\"o\u{g}\"u\c{s} et al. (2000) confirmed these
90: similarities for SGR 1806$-$20 using a larger sample of 401 bursts 
91: (111 detected with the Burst and Transient Source Experiment, BATSE,
92: aboard the Compton Gamma Ray Observatory, CGRO; 290 with the
93: the Proportional Counter Array, PCA, on the Rossi X-ray Timing Explorer, RXTE).
94: The same similarities were found for SGR 1900+14 when we analyzed
95: a sample of 1024 bursts (187 observed with CGRO/BATSE; 837 with the RXTE/PCA) 
96: (G\"o\u{g}\"u\c{s} et al. 1999).
97: Furthermore, we reported evidence for a correlation between duration
98: and fluence for the SGR 1900+14 bursts, similar to the correlation seen 
99: between
100: the duration of strong ground motion at short distances from an earthquake 
101: region and the seismic moment ($\propto$ energy) of an earthquake
102: (Lay \& Wallace 1995). This similarity strongly suggests that SGR bursts, like
103: earthquakes, may be manifestations of self-organized critical systems 
104: (Bak, Tang \& Wiesenfeld 1988), lending support to the hypothesis
105: that SGR bursts involve the release of some form of stored potential
106: energy.
107: 
108: Statistical studies such as the above provide important
109: clues not only about the mechanism by which energy is injected
110: during an SGR burst, but also the mechanism by which it is radiated.
111: For example, in the magnetar model the stored potential energy
112: may be predominantly magnetic, and the electrical currents
113: induced by crustal fractures may be
114: strongly localized over small patches of the star's surface, 
115: somewhat as they are in solar
116: flares.  In such a situation, the energy density in the non-potential
117: magnetic field is high enough that excitation of high
118: frequency Alfv\'en motions leads to rapid damping and the creation
119: of a trapped fireball (through a cascade to high wavenumber:
120: Thompson \& Blaes 1998).  By contrast, if energy is injected more
121: globally, then an approximate balance between injection and radiation
122: will occur in the bursts of luminosity less than $\sim 10^{42}$ erg
123: s$^{-1}$ (Thompson et al. 2000).
124: 
125: Here, we investigate detailed temporal characteristics of a large subset 
126: of SGR 1806$-$20
127: and SGR 1900+14 bursts observed with the RXTE/PCA during their burst active
128: periods in 1996 and 1998, respectively. 
129: We apply to the SGR bursts some temporal analysis methods that were
130: originally developed for the study of GRBs.
131: We also study the spectral variations of the bursts as a function of
132: the burst fluence and duration. In \S~ 2, we briefly review the RXTE/PCA
133: observations. Section 3 describes the data analysis techniques and our 
134: results are discussed in \S~ 4.
135: 
136: 
137: \section{Observations}
138: 
139: The PCA instrument (Jahoda et al. 1996)
140: consists of five Xe proportional counter units (PCUs) sensitive to
141: energies between 2$-$60 keV with 18\% energy resolution at 6 keV. The PCA has a 
142: total effective area of $\sim$ 6700 cm$^{2}$. 
143: 
144: {\bf SGR 1806$-$20 : }
145: The RXTE/PCA observations of SGR 1806$-$20 were performed during a burst active
146: period of the source in 1996 (between November 5 and 18) for a total effective 
147: exposure time
148: of 136.8 ks. Using the burst search procedure described in 
149: G\"o\u{g}\"u\c{s} et al. (2000), we have identified 290 bursts from the source.
150: The number of integrated counts (2$-$60 keV) for these bursts range 
151: between 22 and 34550. Using the count-to-energy conversion factor
152: given in G\"o\u{g}\"u\c{s} et al. (2000) this corresponds to burst fluences
153: between $1.2 \times 10^{-10}$ and $1.9 \times 10^{-7}$ ergs cm$^{-2}$ 
154: (E $>$ 25 keV).
155: Assuming isotropic burst emission, the corresponding energy range is
156: $3.0 \times 10^{36}$ -- $4.9 \times 10^{39}$ ergs, for a distance to 
157: SGR 1806$-$20 of 14.5 kpc (Corbel et al. 1997).
158: To facilitate our analysis, we chose events that were clustered together during
159: two very active epochs of the source, resulting in 268 events recorded on
160: 5 (MJD 50392) and 18 (MJD 50405) November 1996.
161: 
162: {\bf SGR 1900+14 : }
163: RXTE observations of SGR 1900+14 took place
164: between 1998 June 2 and December 21, for a total effective exposure time of
165: 224.1 ks. Using the same burst search algorithm as before, we identified a 
166: total of 837 bursts from the source with integrated counts
167: ranging between 22 and 60550. The fluences of these bursts range 
168: from $1.2 \times 10^{-10}$ to $3.3 \times 10^{-7}$ ergs cm$^{-2}$ (E $>$ 25 keV) 
169: corresponding to an energy range of $7 \times 10^{35}$ -- $2 \times 10^{39}$ 
170: ergs (assuming isotropic emission at 7 kpc [Vasisht et al.~1994]). 
171: Similarly, we selected 679 events which occurred during a very active period of
172: the source between 1998 August 29 (MJD 51054) and September 2 (MJD 51058).
173: 
174: 
175: \section{Data Analysis and Results}
176: 
177: \subsection{Duration (T$_{90}$) estimates}
178: 
179: Originally defined for cosmic GRBs, the T$_{90}$ duration of a burst is the time
180: during which 90\% of the total (background-subtracted) burst counts have been
181: accumulated since the burst trigger (Kouveliotou et al. 1993). 
182: We calculated the 
183: T$_{90}$ duration of SGR bursts using event-mode PCA data (2-60 keV)
184: with 1/1024 s time resolution. For each burst we collected the cumulative counts
185: for an 8 s continuous stretch of data starting 4 s before its peak, t$_{\rm p}$.
186: We then fit the cumulative count distribution 
187: between two user-selected background intervals\setcounter{footnote}{5}
188: \footnote{Approximately equal background intervals were selected before and
189: after the burst, except in cases where separation between bursts was too small;
190: there we chose smaller, but significant, post/pre burst intervals.} 
191: with a first order polynomial
192: plus a step function (for the burst) and subtracted the background 
193: counts. By using a first order polynomial to fit the cumulative counts, we 
194: assume the PCA background remains flat over each 8 s segment. The resulting
195: height of the step function gives the total burst counts. Figure 1 depicts the
196: steps of the T$_{90}$ estimate procedure for one of the SGR 1900+14 bursts.
197: 
198: As the selected bursts occurred during extremely active periods 
199: (for both sources), there were quite a number of
200: cases where bursts were clustered very close together.
201: During these active episodes, many crustal sites on the neutron star
202: may be active, releasing stored potential energy in the form of bursts
203: with a large variety of time profiles.
204: Hence, it is important to distinguish single pulse events from events 
205: with multiple peaks (which may involve multiple fracture sites). 
206: To do this, we applied arbritrary but consistent criteria to our data.
207: We classified an event as {\it multi-peaked} if the count rate at any local
208: minimum (in 7 ms time bins) is less than half the maximum value attained 
209: subsequently in the burst (see Figure 2, middle plots).  
210: Otherwise the event was classified as {\it single-peaked} (Figure 2, 
211: top plots).  When the count rate dropped to the noise level between peaks, 
212: the event was classified as a single, multi-peaked burst if and only if the 
213: time between peaks was less than a quarter of the neutron star rotation 
214: period ( 1.3 s for SGR 1900+14, 1.9 s for SGR 1806-20; see Figure 2, bottom 
215: plots).  We found that 262 of the 679 bursts from SGR 1900+14, and 113 of 
216: the 268 SGR 1806-20 bursts were multi-peaked by these criteria.   
217: Note that the total counts per burst (or "count fluences") were found
218: to span nearly the same range in the two classes of bursts, in both 
219: sources.  This suggests that the light curve peak structure 
220: does not correlate strongly with total energy.
221: 
222: In the context of cosmic GRBs, the value of T$_{90}$ can be systematically 
223: underestimated in faint events due to low signal-to-noise (Koshut et al. 1996). 
224: We investigated this effect in our SGR analysis 
225: using extensive numerical simulations. We created three time profiles based 
226: upon the observed time profiles; a single two-sided 
227: Gaussian (with right-width twice the left width), two two-sided Gaussians whose 
228: peaks are separated by 0.5 s and the peak rate of the second pulse is 0.6 of 
229: the first one,
230: and two two-sided Gaussians with peak separation of 1 s. For each profile, we
231: varied the peak rates (eight values between 2600 counts s$^{-1}$ and 55000 
232: counts s$^{-1}$)
233: and the width of the Gaussians (eight values from 12 ms to 120 ms). 
234: We determined 
235: the T$_{90}$ duration for each combination before adding noise. Then for each
236: combination, we generated 800 realizations including Poisson noise and
237: determined the respective T$_{90}$ values for these simulations. We found
238: that the fractional difference, FD = 1 $-$ (T$_{90,\rm s}$ / T$_{90,\rm a}$)
239: between the actual value, T$_{90,\rm a}$, 
240: (in the absence of noise) and the simulated one, T$_{90,\rm s}$, has a 
241: strong dependence on the peak rate (i.e. signal-to-noise ratio), 
242: such that FD can be as high as 0.22 at count
243: rates of 2600 counts s$^{-1}$. 
244: It is, however, less than 0.05 for count rates greater
245: than 4800 counts s$^{-1}$. The variation of FD with respect to 
246: the pulse profile is insignificant.   
247: Using the results of our simulations we obtained a 
248: T$_{90}$ correction as a function of peak rate.
249: We estimated the peak
250: rate of each burst using a box-car averaging technique with a box width of 
251: 1/512 s.
252: We then corrected T$_{90}$ values for the S/N effect using our correction 
253: function and these peak rates.
254: 
255: Our final data set thus comprises 455 SGR 1900+14 bursts with corrected
256: $T_{90}$ durations between 9 ms and 2.36 ms (Figure 3, solid histogram).
257: We were unable to determine statistically significant $T_{90}$ durations 
258: for 187 SGR 1900+14 bursts because there were less than $\sim 40$
259: counts per event.
260: The solid curve in Figure 3 is a best fit log-Gaussian 
261: function to all T$_{90}$ values which peaks at 93.9 $\pm$ 0.2 ms 
262: ($\sigma$ = 0.35 $\pm$ 0.01; where $\sigma$ is the width of the distribution 
263: in decades). The dashed histogram in Figure 3 is the distribution of 
264: T$_{90}$ values of single pulse bursts whose log-Gaussian mean is 
265: 46.7 $\pm$ 0.1 ms ($\sigma$ = 0.21 $\pm$ 0.01). Also in Figure 3, 
266: the dash-dot histogram displays the T$_{90}$ distribution of multi-peaked
267: bursts and it peaks at 148.9 $\pm$ 0.2 ms ($\sigma$ = 0.26 $\pm$ 0.02).
268: For SGR 1806$-$20 bursts, we determined corrected T$_{90}$ values of 190 bursts
269: which range between 16 ms and 1.82 s (49 bursts were too weak to determine 
270: their T$_{90}$ values). 
271: The distribution of all T$_{90}$ values
272: is shown in Figure 4 (solid histogram). A log-Gaussian fit to this distribution
273: yields a peak at 161.8 $\pm$ 0.2 ms ($\sigma$ = 0.34 $\pm$ 0.02).
274: Similarly, a log-Gaussian fit to the
275: T$_{90}$ distribution of the single pulse bursts (dashed lines in Figure 4) 
276: peaks at 88.1 $\pm$ 0.1 ms ($\sigma$ = 0.19 $\pm$ 0.03) and a fit to the 
277: multi-peaked bursts (dash-dot lines in Figure 4) yields a peak at
278: 229.9 $\pm$ 0.3 ms ($\sigma$ = 0.32 $\pm$ 0.03).
279: 
280: In order to quantify the time profile symmetry of SGR bursts we determined the 
281: ratio of their rise times (t$_{r}$, i.e. the interval between T$_{90}$ start 
282: time and peak time t$_{p}$) to T$_{90}$ durations of the single pulse events. 
283: In Figure 5, we plot the distributions of these ratios for bursts with single 
284: pulse (solid histogram) and milti-peaks (dashed histogram) from 
285: SGR 1900+14 (left) and those from SGR 1806$-$20 (right). 
286: The majority of single pulse events from both sources have t$_{r}$ / T$_{90}$
287: values less than 0.3, showing that SGR bursts decay slower than they rise, 
288: or in other words that SGR pulse profiles are asymmetric. We find that the 
289: average values os the t$_{r}$ / T$_{90}$ ratios are 0.29 and 0.27 for 
290: SGR 1900+14 and 
291: SGR 1806$-$20, respectively. Remarkably coincident average values, along with
292: very similar distributions of t$_{r}$ / T$_{90}$ ratios of both sources suggest
293: a similarity of the asymmetry in the temporal profiles of SGR bursts.  
294: 
295: 
296: \subsection{Emission time ($\tau$$_{90}$) estimates}
297: 
298: Emission time, $\tau$$_{N}$, was introduced for cosmic GRBs as a 
299: complimentary temporal parameter to T$_{90}$
300: (Mitrofanov et al. 1999). Emission time is the time
301: over which a fixed percentage, N\%, of the total burst emission was
302: recorded, starting from the peak of the event and moving downward in flux.
303: For each burst, we determined the average background level from the background
304: intervals used in the T$_{90}$ estimating procedure. The
305: background-subtracted count bins within the burst interval (i.e. from
306: the end of pre-burst background range to the beginning of post-burst background 
307: range) were then ordered by decreasing count rate. 
308: Starting with the highest count rate bin (i.e. the peak), 
309: we added the counts of each successively weaker bin until 90\%
310: of the total burst counts were accumulated. 
311: The $\tau$$_{90}$ emission time of the
312: burst is then the total time spanned by the accumulated bins. 
313: 
314: As with the T$_{90}$ parameter, when the S/N ratio of the burst is
315: low, the measured emission time can be systematically smaller than the 
316: actual value (Mitrofanov et al. 1999). In order to correct for this
317: systematic error, we constructed numerous simulated profiles identical to those 
318: described in the previous section.
319: We found that, similar to T$_{90}$ estimates, the FD 
320: between the actual and simulated $\tau$$_{90}$ emission times is strongly
321: dependent on the peak rate. FD is $\sim$ 0.26 at count rates of 2600 
322: counts s$^{-1}$
323: and less than 0.06 for count rates greater than 4800 counts s$^{-1}$.
324: Similar to the T$_{90}$ measurement, this effect is only weakly dependent on the
325: temporal profile of the burst.
326: Using our $\tau$$_{90}$ correction function and the peak rates,
327: we obtained the corrected $\tau$$_{90}$ emission times. The corrected values
328: range between 4.6 and 412.8 ms for SGR 1900+14 bursts, and between 
329: 4.8 and 559.2 ms for SGR 1806$-$20 bursts.
330: In Figure 6, we show the distributions of $\tau$$_{90}$ emission times for
331: SGR 1900+14 ({\it left}) and SGR 1806$-$20 ({\it right}). The dashed lines in
332: both plots are the best fit log-Gaussian curves which peak at 49.6 $\pm$ 0.1 ms
333: ($\sigma$ = 0.28 $\pm$ 0.01) for SGR 1900+14 events, and at 82.3 $\pm$ 0.1 ms 
334: ($\sigma$ = 0.32 $\pm$ 0.02) for SGR 1806$-$20 events.
335: The $\tau$$_{90}$ emission times of bursts from both sources are on average 
336: shorter than T$_{90}$ durations. Note that the log-Gaussian
337: mean values of T$_{90}$ distributions of {\it single} pulse bursts from both 
338: SGRs are quite similar to those of the $\tau$$_{90}$ distributions.
339: 
340: 
341: \subsection{Duty cycles ($\delta$$_{90}$)}
342: 
343: As shown in the previous section, the $\tau$$_{90}$ values of bursts are in 
344: general smaller than their T$_{90}$ values. 
345: The reason is that $\tau$$_{90}$ is not a
346: measure of the actual duration of the burst when the morphology of the burst is
347: complex (e.g. multiple peaks). 
348: Mitrofanov et al. (1999) suggested that the ratio $\tau$$_{90}$/T$_{90}$ 
349: can then be used to describe a duty cycle, $\delta$$_{90}$. 
350: We, therefore, determined the $\delta$$_{90}$ parameters for all bursts of both
351: sources (Figure 7, {\it left} for SGR 1900+14 bursts, {\it right} for SGR
352: 1806$-$20 bursts). Both distributions peak at about $\delta$$_{90}$ $\sim$ 0.45
353: and their overall shapes are very similar, indicating a strong similarity of the
354: {\it type} of burst emission in the two sources, although their overall duration
355: distribution differ. 
356: 
357: 
358: \subsection{Duration -- Fluence -- Hardness correlations}
359: 
360: Gutenberg and Richter (1956) showed that there exists a power law 
361: relation between the magnitude (which is related to the total energy involved) 
362: of earthquakes and the durations of strong ground-shaking
363: at short distances from an earthquake epicenter. We investigated whether a 
364: similar
365: correlation exists in SGR events using their total burst counts and their 
366: T$_{90}$ durations (as estimated in \S 3.1), 
367: for both SGR 1806$-$20 and SGR 1900+14.
368: For each set of bursts, we grouped the T$_{90}$ values
369: into logarithmically spaced bins and determined the weighted mean 
370: value of total burst counts and T$_{90}$ durations for each bin. 
371: In Figure 8, we show the plot of integral counts vs T$_{90}$ durations
372: of SGR 1806$-$20 bursts. The crosses in this figure are the weighted means of
373: each parameter; the errors on the mean-T$_{90}$ values denote the range of 
374: each bin while the errors on the mean counts are 
375: due to sample variance. The dark points are the
376: individual measurements of the burst counts and T$_{90}$ values of single pulse
377: events and the gray circles are those of multi-peaked bursts.
378: Figure 8 shows that the integral counts and durations of SGR 1806$-$20 bursts 
379: are well correlated. To quantify this correlation, we determined the 
380: Spearman's rank order correlation coefficient, $\rho$ = 0.91, and the
381: probability of getting this value from a random data set, P=3.4 $\times$
382: 10$^{-4}$. We further fit a power law model to the mean values (crosses) of the
383: data using the least squares technique, which yields a power law index of 
384: 1.05 $\pm$ 0.16.
385: 
386: Similarly, Figure 9 shows that the integral counts and T$_{90}$ durations of 
387: SGR 1900+14 bursts are also correlated, having 
388: $\rho$ = 0.89 and P = 8.6 $\times$ 10$^{-4}$. A power law fit
389: to the mean values of the data yields an index of 0.91 $\pm$ 0.07. The asterisk
390: shown in the upper right portion of Figure 9 indicates where the precursor of
391: the August 29 burst falls. This event was exceptionally long, bright and 
392: resembled 
393: the August 27 giant flare in various ways (Ibrahim et al. 2000).
394: A spectral line at 6.4 keV was reported during the precursor of event of
395: August 29 burst with $\sim$ 4 $\sigma$ significance 
396: (Strohmayer \& Ibrahim 2000). 
397: If spectral lines are event intensity dependent, there are very few events
398: in Figure 9 during which we would expect to see any lines. The detailed 
399: spectral analysis of all SGR bursts is underway.
400: 
401: In conclusion we find a good correlation between the duration and total counts
402: (energy) of SGR bursts, quite similar to the one established for earthquakes.
403: It is noteworthy that the burst count fluences of single pulse events (from both 
404: systems) span a range almost as wide as that of fluences of multi-peaked 
405: bursts.   
406: 
407: In order to investigate burst spectral variations versus 
408: fluence, we calculated an event hardness ratio 
409: defined as the ratio of the total burst counts observed between 10$-$60 keV 
410: to those between 2 $-$ 10 keV.
411: Bursts with total counts less than $\sim$ 50 yielded
412: statistically insignificant hardness ratios ($<$ 3 $\sigma$), and therefore,
413: were excluded from our analysis. 
414: For SGR 1806$-$20 bursts, we have 159 events with hardness ratios
415: measured to $>$ 3 $\sigma$ accuracy.
416: We divided the total counts of these events into logarithmically
417: spaced bins and determined the weighted mean
418: hardness ratio for each group. Figure 10a shows that the SGR 1806$-$20
419: burst hardness ratios are anti-correlated with fluence 
420: ($\rho$=$-$0.96, P=2.6 $\times$ 10$^{-4}$).  
421: For 385 SGR 1900+14 bursts, Figure 10d shows a marginal 
422: anti-correlation between hardness and fluence
423: ($\rho$ = $-$0.89, P=4.6 $\times$ 10$^{-3}$).
424: Interestingly, we find that although the hardness $-$ fluence anti-correlation
425: is evident in both sources, the SGR 1806$-$20 events are {\it overall} harder
426: than the SGR 1900+14 ones. Our energy selection optimizes the PCA energy
427: response, so that we are not affected by instrumental biases.
428: 
429: Fenimore, Laros \& Ulmer (1994) have performed a similar study of hardness 
430: ratio vs
431: fluence for 95 SGR 1806-20 events detected with ICE. We cannot directly compare
432: our results with this study for two reasons. The energy ranges over which 
433: hardness ratios are computed differ significantly; while we use 
434: (10 $-$ 60 keV)/(2 $-$ 10 keV), Fenimore et al. (1994) use 
435: (43.2 $-$ 77.5 keV)/(25.9 $-$ 43.2 keV). The PCA sensitivity drops significantly
436: beyond $~$ 25 keV, limiting the possibility of comparison.
437: Furthermore, our fluence range ends at $~2 \times 10^{-7}$ ergs cm$^{-2}$, 
438: just below the range described in Figure 1 of Fenimore et al. (1994).
439: However, since Fenimore et al. report a constant trend above 
440: $10^{-7}$ ergs cm$^{-2}$, one could postulate that the hardness
441: ratio vs fluence trend we report, levels off at higher fluences (energies).
442: 
443: We further investigated the hardness $-$ fluence trend for burst morphology
444: sub-sets, namely for single pulse and multi-peaked burst groups for each 
445: source. 
446: Figure 10b exhibits hardness $-$ fluence plot of single pulse SGR 1806$-$20 
447: events and Figure 10c shows that of multi-peaked bursts.
448: We see that both sets display spectral softening as the burst count fluence
449: increases. Similarly SGR 1900+14 events are shown in Figure 10e and f.
450: For this source, the hardness $-$ fluence anti-correlation is significant only
451: for single pulse bursts. 
452: It is important to note that the peak rates of most of the 
453: highest fluence bursts reach $\gtrsim$ 10$^5$ counts s$^{-1}$ (on 1/1024 s 
454: time scale) around which the PCA pulse pileup may become important. This effect 
455: can
456: artificially harden the observed count spectrum at these rates. The spectral
457: hardening or leveling off seen in the last bins of the plots of Figure 10
458: may well be due to the pulse pileup effect and should not be considered as
459: intrinsic source property.
460:  
461: We next investigated the relationship between the hardness and duration of SGR 
462: bursts. Although we find hardness $-$ fluence anti-correlations and
463: fluence $-$ duration correlations, both SGR 1806$-$20 (Figure 11, squares) and
464: SGR 1900+14 (Figure 11, diamonds) burst hardness ratios are independent of 
465: the event durations.
466: 
467: 
468: \section{Discussion}
469: 
470: Our study demonstrates that unlike the T$_{90}$ duration distribution of cosmic
471: GRBs which shows a bi-modal trend with peaks at $\sim$ 0.31 s and $\sim$ 37 s
472: (Kouveliotou et al. 1993; Paciesas et al. 1999), the T$_{90}$ distribution of 
473: SGR bursts displays a single peak which varies for each source: $\sim$ 93 ms
474: and $\sim$ 162 ms for SGR 1900+14 and SGR 1806$-$20, respectively. 
475: 
476: We find that the ${\rm T}_{90}$ durations of single-pulse bursts
477: from both SGRs form narrow distributions (compared
478: to those of multi-peaked events) which peak at $\sim 47$ ms
479: and $\sim 88$ ms for SGR 1900+14 and SGR 1806-20, respectively.
480: These bright, hard, and short bursts are almost certainly powered
481: by a sudden disturbance of the rigid neutron star crust, which
482: transmits energy to the magnetosphere.  Magnetic stresses,
483: which force the star between distinct metastable equilibria,
484: provide the most plausible source of energy for the giant flares,
485: and by inference for the short bursts (TD95; Thompson \& Duncan 2001).
486: 
487: The duration of the multi-peaked bursts is clearly fixed by the 
488: time between successive releases of energy.  The widths of the
489: single pulse bursts could, in principle, be limited either
490: by the rate of release of energy from the initial reservoir 
491: or, alternatively, by the time for the released energy to be
492: converted to radiation through some intermediate reservoir.   There have
493: been various suggestions for such an intermediate storage mechanism:
494: a hot fireball that is confined on closed magnetic field lines 
495: (TD95);   a region of strong magnetic shear and 
496: high current density (Thompson et al. 2000); or a persistent
497: vibration of the star (Fatuzzo \& Melia 1994). 
498: 
499: Measurement of the burst rise time $t_r$ provides a discriminant 
500: between these possibilities.  We find that $t_r$ is characteristically much
501: shorter than the total duration, as defined by either ${\rm T}_{90}$
502: or $\tau_{90}$. Moreover, the distribution of $t_r/{\rm T}_{90}$ is broad
503: (both for single-pulsed bursts and multi-peaked events),
504: which suggests that $t_r$ is not directly connected to cooling.
505: For example, the X-ray luminosity of a trapped fireball in local
506: thermodynamic equilibrium is proportional to its surface area, and
507: is a weak function of its internal temperature (Thompson \& Duncan 1995).
508: The characteristic radiative timescales are, in general, very short
509: at the high spectral intensity of an SGR burst. 
510: Thus, the rise of the X-ray flux plausibly represents
511: the initial injection of energy, although strictly $t_r$ only sets an 
512: upper bound to the injection timescale.
513: 
514: 
515: While the overall shape of T$_{90}$ (both single and multi-peaked) and 
516: the $\tau$$_{90}$ distributions of 
517: both sources are quite similar (relatively consistent Gaussian
518: widths), the peaks of both distributions for SGR 1900+14 occur at shorter
519: durations compared to those for SGR 1806$-$20. 
520: This systematic difference in the burst durations probably results
521: from some differing intrinsic property of the sources, such as the
522: strength of the magnetic field, or the size of the active region.
523: 
524: We find a power law correlation between the total burst counts (fluence) 
525: and the duration of SGR bursts with a power law index around 1. Similar 
526: behavior was noted for earthquakes by Gutenberg \& Richter (1956).
527: Within the context of earthquake mechanics, one way of defining duration
528: (also known as ``bracketed duration" [Bolt 1973]) is the time between the 
529: first and last 5\% excesses of g (gravitational acceleration on Earth) by 
530: the threshold acceleration of strong ground shaking. Recently, Lay \& Wallace 
531: (1995) presented the power law correlation between the seismic moment 
532: ($\propto$ energy) and duration of 122 earthquakes with energies between 
533: 3.5 $\times$ 10$^{23}$ erg and 2.8 $\times$ 10$^{26}$ erg. The power law 
534: fit to these events yields an index of 3.03.  
535: 
536: An equally important constraint on the injection and cooling mechanisms
537: comes from the anti-correlation between the hardness and fluence of the
538: SGR bursts.  Although very significant for SGR 
539: 1806-20, this anti-correlation is much milder than that expected for 
540: black-body emission from a region of constant area; indeed,
541: the trend of increasing hardness with lower fluence 
542: is opposite to that expected for constant area emission.  
543: Two basic types of radiative mechanism could reproduce this trend.
544: First, the emitting plasma could be in local thermodynamic equilibrium, 
545: which requires that its size (radiative area) should decrease at lower
546: fluences.  An alternative possibility is that the spectral intensity 
547: of the radiation field sits below that of a black body, and that the
548: temperature of the emitting plasma is buffered within a narrow range.
549: We consider each of these possibilities in turn.
550: 
551: An SGR burst can be parameterized by the rate of injection of 
552: energy into the magnetosphere, $L_{\rm inj}$, and the volume $V$
553: of the injection region.  When $L_{\rm inj}$ is large and $V$ is
554: small enough, it is not possible to maintain a steady balance
555: between heating and radiative cooling.  The deposited energy is
556: locked onto closed magnetic field lines of the neutron star, in
557: a ``trapped fireball'' composed of photons and electron-positron pairs
558: (TD95).  This kind of event will tend to have a soft spectrum, because 
559: the injected energy has thermalized, and the plasma remains in LTE
560: very close to its photosphere.  The rise time is comparable to the
561: time over which energy is initially injected, but the decay is
562: limited by the rate of cooling through a thin radiative surface layer,
563: which contracts toward the center of the fireball.  The declining light
564: curve of the 27 August 1998 giant flare can be accurately fit by such
565: a model (Feroci et al. 2001;  Thompson \& Duncan 2001).  
566: 
567: A second burst from SGR 1900+14 on 29 August 1998 has been interpreted
568: in this trapped fireball model (Ibrahim et al. 2000).  The main 29 August burst
569: had a much shorter duration than the giant flare ($\sim 3.5$ s versus
570: $\sim 400$ s).  This bright component was followed by a much fainter 
571: pulsating tail (extending out to $\sim 1000$ s) which
572: provides direct evidence for heating and compression of the neutron
573: star surface by the fireball.  In this model, the short duration and 
574: high luminosity of the bright component require that the trapped fireball
575: had an approximately planar geometry, as would be expected if the energy
576: were released along an extended fault (Ibrahim et al. 2000).  
577: 
578: As our results show, most single-pulsed bursts from SGR 1900+14 have a 
579: much shorter duration (40 times smaller) than the bright component of 
580: the August 29 burst.  It remains unclear, therefore, whether
581: the trapped fireball model also applies to these much more frequent
582: events.  If it does apply, then the narrowness of the ${\rm T}_{90}$
583: distribution (compared with the wide range of measured fluences) also
584: requires a planar geometry, because the cooling time is determined
585: by the smallest dimension of the fireball.
586: 
587: The radiative mechanism is somewhat different if the injection luminosity
588: $L_{\rm inj}$ lies below a critical value of $\sim 10^{42}\,(V^{1/3}/
589: 10~{\rm km})$ ergs s$^{-1}$ (assuming a spherical geometry;  
590: Thompson \& Duncan 2001).
591: When the compactness is that low,  it is possible to maintain a steady
592: balance between the heating of a corona of electron-positron
593: pairs, and radiative diffusion out of the corona (which one deduces must
594: be optically thick to scattering).  The central temperature of the corona 
595: is buffered in the range $\sim 20-40$ keV, and remains higher at
596: {\it lower} luminosities -- just the trend observed for SGR 1806-20 (Fig. 10).
597: If this radiative model applies to the majority of single-pulsed
598: events, then the constraints on the geometry of the active region are
599: weaker, and one must consider alternative mechanisms for
600: storing the injected energy, such as a persistent current driven
601: by shearing motions in the neutron star crust (Thompson et al. 2000).
602: 
603: 
604: \acknowledgments
605: 
606: 
607: E.G. is grateful to James C. Pechmann for useful discussions on earthquakes.
608: We acknowledge support from NASA grants NAG5-6021, NAG5-7785(E.G); 
609: the LTSA grant NAG 5-9350 (P.M.W. \& C.K.);
610: NASA grant NAG5-3100 and Alfred P. Sloan foundation (C.T.);
611: the Texas Advanced Research Project grant ARP-028 and 
612: NASA grant NAG5-8381 (R.C.D.). Several of the ideas presented in this work
613: originated at the ITP funded by NSF grant PHY99-07949.
614: 
615: 
616: \begin{references}
617: 
618: \reference{bak88} Bak, P., Tang, C. \& Wiesenfeld, K. 1988, \pra, 38, 364
619: \reference{blt73} Bolt, B.A. 1973, in 5th World Conf. Earthquake Eng., 
620: Inter. Assoc. for Earthquake Eng., Rome, paper no. 292
621: \reference{cheng95} Cheng, B., et al. 1996, \nat, 382, 518
622: \reference{cl81} Cline, T.L., et al. 1981, \apjl, 255, L45
623: \reference{cor97} Corbel, S., et al. 1997, \apj, 478, 624
624: \reference{dt92} Duncan, R.C. \& Thompson C. 1992 \apjl, 392, L9
625: \reference{fa94} Fatuzzo, F, \& Melia, F. 1994, in AIP Proc., 2th Huntsville 
626: Symp. on
627: Gamma-Ray Bursts, eds. G.J. Fishman, J.J. Breinerd \& K. Hurley (New York:
628: AIP), p.515
629: \reference{fe94} Fenimore, E.E., Laros, J.G. \& Ulmer, A. 1994, \apj, 432, 742 
630: \reference{fe99} Feroci, M., et al. 1999, \apjl, 515, L9 
631: \reference{fe01} Feroci, M., Hurley, K., Duncan, R.C., \& Thompson C. 2001,
632: \apj, in press, astro-ph/0010494
633: \reference{go99} {G\"o\u{g}\"u\c{s}}, E., et al. 1999, \apjl, 526, L93
634: \reference{go00} {G\"o\u{g}\"u\c{s}}, E., et al. 2000, \apjl, 532, L121
635: \reference{gr56} Gutenberg, B. \& Richter, C.F. 1956, Bull. Seis. Soc.
636: Am., 46, 105
637: \reference{hur99} Hurley, K., et al. 1999, \nat, 397, 41
638: \reference{hur00} Hurley, K., et al. 2000, in AIP Proc., 5th Huntsville Symp. on
639: Gamma-Ray Bursts, eds. R.M. Kippen, R.S. Mallozzi \& G.J. Fishman (New York:
640: AIP), p.763
641: \reference{jad96} Jahoda, K. et al. 1996, Proc. SPIE, 2808, 59 
642: \reference{ibr00} Ibrahim, A., et al. 2001, \apj, submitted, astro-ph/0007043 
643: \reference{kou93} Kouveliotou, C., et al. 1993, \apjl, 413, L101
644: \reference{kos96} Koshut, T.M., et al. 1996, \apj, 463, 570
645: \reference{lay95} Lay, T. \& Wallace, T., eds. 1995, Modern Global Seismology,
646: (San Diego: Academic), p.378
647: \reference{maz79} Mazetz, E.P., et al. 1979, Nature, 282, 587
648: \reference{maz99} Mazetz, E.P., et al. 1999, Astronomy Letters, 25, 635 
649: \reference{mit99} Mitrofanov, I.G., et al. 1999, ApJ, 504, 925
650: \reference{pcs92} Paciesas, W.S., et al. 1999, \apjs, 122, 465
651: \reference{pac92} Pac{\` z}ynski, B. 1992, Acta Astron., 42, 145
652: \reference{str00} Strohmayer, T \& Ibrahim, A. 2000, \apjl, 537, 111
653: \reference{td95} Thompson, C. \& Duncan, R.C. 1995, \mnras, 275, 255
654: \reference{td96} Thompson, C. \& Duncan, R.C. 1996, \apj, 473, 322
655: \reference{tb98} Thompson, C. \& Blaes, O. 1998, Phys. Rev. D, 57, 3219
656: \reference{td00} Thompson, C., et al. 2000, \apj, 543, 340
657: \reference{td01} Thompson, C. \& Duncan, R.C. 2001, \apj, submitted
658: \reference{vas94} Vasisht, G., et al. 1994, \apj, 431, L35
659: 
660: \end{references}
661: 
662: \newpage
663: 
664: \begin{figure}
665: \plotone{f1.eps}
666: \caption{({\it top panel}) Light curve of a SGR 1900+14 burst (2$-$60 keV) 
667: accumulated with 20 ms time resolution. The vertical dot-dash lines before and
668: after the burst are the boundaries of the selected pre- and post-background
669: intervals. ({\it middle panel}) The solid line shows the cumulative counts, 
670: with the background
671: contribution (as fit by a first-order polynomial) subtracted off.  The 
672: dashed line shows a step-function fit to the burst counts.
673: The lower horizontal dotted line represents the 5\% and upper horizontal 
674: dotted shows the 95\% count fluence level. Note that the top and the middle
675: panels are in the same time scale. ({\it bottom panel}) 
676: Time profile of the same burst
677: with 7 ms time resolution and zoomed in near the time of the burst. 
678: The vertical dotted lines
679: are the start and the end times of its T$_{90}$ duration.} 
680: \end{figure}
681: 
682: \newpage
683: 
684: \begin{figure}
685: \plotone{f2.eps}
686: \caption{Time profiles of some SGR 1900+14 and SGR 1806$-$20 bursts (on 7 ms
687: time scale) to illustrate single pulse bursts (top plots) and multi-peaked 
688: bursts (middle and bottom plots).}
689: \end{figure}
690: 
691: \newpage
692: 
693: \begin{figure}
694: \plotone{f3.eps}
695: \caption{Distributions of the T$_{90}$-durations for all (solid), single pulse 
696: (dashed) and multi-peaked (dash-dot) SGR 1900+14 bursts. 
697: The solid, dashed and dash-dot curves are obtained by fitting 
698: a log-Gaussian model to each set. Sets peak at 93.4 ms, 46.7 ms and 148.9 ms, 
699: respectively.}
700: \end{figure}
701: 
702: \newpage
703: 
704: \begin{figure}
705: \plotone{f4.eps}
706: \caption{Distributions of the T$_{90}$-durations for all (solid), single pulse 
707: (dashed) and multi-peaked (dash-dot) SGR 1806$-$20 bursts. 
708: The solid, dashed and dash-dot curves are obtained by fitting 
709: a log-Gaussian model to each set. Sets peak at 161.8 ms, 88.1 ms and 229.9 ms, 
710: respectively.} 
711: \end{figure}
712: 
713: \newpage
714: 
715: \begin{figure}
716: \plotone{f5.eps}
717: \caption{Distributions of the ratios of rise times, t$_{r}$ to T$_{90}$ 
718: durations for single pulse bursts (solid) and multi-peaked bursts (dashed)
719: from SGR 1900+14 (left) and SGR 1806$-$20 (right). Note that the number of
720: multi-peaked events with the lowest t$_{r}$/T$_{90}$ ratios is artificially 
721: large due to the longer (compared to t$_{r}$) durations of these bursts 
722: (see Figure 2, middle and bottom plots).}
723: \end{figure}
724: 
725: \newpage
726: 
727: \begin{figure}
728: \plotone{f6.eps}
729: \caption{Distributions of the $\tau$$_{90}$ emission times for SGR 1900+14 
730: (left) and SGR 1806$-$20 (right) bursts. The dashed curves are obtained by 
731: fitting a
732: log-Gaussian model to each set, with peaks at 49.6 ms and 82.3 ms, 
733: respectively.}
734: \end{figure}
735: 
736: \newpage
737: 
738: \begin{figure}
739: \plotone{f7.eps}
740: \caption{Distributions of the duty cycles, $\delta$$_{90}$, for SGR 1900+14 
741: (left) and SGR 1806$-$20 (right) bursts. The Gaussian fits peak at 0.45 and 0.46,
742: respectively.}
743: \end{figure}
744: 
745: \newpage
746: 
747: \begin{figure}
748: \plotone{f8.eps}
749: \caption{Plot of the total counts vs T$_{90}$-durations for 190 SGR 1806$-$
750: 20 bursts which shows a correlation ($\rho$ = 0.91, P=3.4 $\times$ 10$^{-4}$). 
751: The solid line is obtained by fitting a power law model to the crosses (which
752: are explained in the text). The best fit power law index is 1.05 $\pm$ 0.16.
753: The circles are the individual measurements for single pulse (dark) and 
754: multi-peaked (gray) bursts.}
755: \end{figure}
756: 
757: \newpage
758: 
759: \begin{figure}
760: \plotone{f9.eps}
761: \caption{Plot of the total counts vs the T$_{90}$-durations for 455 SGR 
762: 1900+14 bursts which shows a correlation ($\rho$ = 0.89 and P = 8.6 
763: $\times$ 10$^{-4}$). A power law fit to the crosses (solid line) yields an 
764: index of 0.91 $\pm$ 0.07. The asterisk indicates where August 29 precursor
765: would be seen. The circles represent the measured values for single pulse 
766: (dark) and multi-peaked (gray) bursts.}
767: \end{figure}
768: 
769: \newpage
770: 
771: \begin{figure}
772: \plotone{f10.eps}
773: \caption{Plot of hardness ratios vs total counts for (a) all, (b) single 
774: pulse, (c) multi-peaked SGR 1806$-$20 bursts. Similarly, those for (d) all,
775: (e) single pulse, (f) multi-peaked SGR 1900+14 events.}
776: \end{figure}
777: 
778: \newpage
779: 
780: \begin{figure}
781: \plotone{f11.eps}
782: \caption{Plot of hardness ratios vs T$_{90}$ durations for SGR 1806$-$20 bursts
783: (squares) and for SGR 1900+14 (diamonds).}
784: \end{figure}
785: 
786: 
787: 
788: \end{document}
789: