RHESSI-NESSIE workshop, working group 1 June 4 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ Schwartz, Mrozek, Kontar, Trottet, McKinnon, Krucker, Aschwanden, Gan, Hurford, Dennis, Smith, Hudson (clockwise from above) 1. Introduction (Smith) * Instrument characteristics illustrated by the July 23 event * Pileup illustration from April 21 event * Data dropouts * Detector design OHP * Solicitation of questions from group (no responses) 2. Spectroscopy (Schwartz) * Inside the HSI_SPECTRUM and HSI_LIGHTCURVE objects and SPEXophobia - Gain calibration is _really_ linear, based on pre-flight cals - Question from Kontar about 10-15 keV features that could influence inversions - Rebinning PHA channels to energy bins. Fractional count method. At low energies, channels are ~1/3 keV; expect artefacts. PHA is "equisitely" flat - Lack of low-energy calibration lines; extrapolation from 93 keV is probably not good enough for 3-10 keV work - Discussion of energy sampling of models (Dennis) - Granularity of live-time measures (not so good at lower rates; no better than 512 microsec!). Hurford comments - it's not so important at low rates; at high rates it may resemble an additional weighting term. Need to address this when we are doing harmonics of the modulation (TBD). - Comment (Hurford) regarding pileup artifacts in the imaging - Livetime integration and counts integration not properly aligned yet and could introduce a phase error in imaging at high rates (TBD) - Decimation overview. Sticky points remain... e.g. partial binning. TBD. Illustration of various effects with observing summary page for May 29, 2003 gamma-ray flare. HESSI_DATA program. CHANGES method. - Datagaps overview. Coincidence overview (Smith) - Bug reported in the detector response matrix software at the lowest energies (TBD) 3. Time series (Hudson) * see .ppt file @ http://sprg.ssl.berkeley.edu/~hhudson/presentations/rhessinessi motivating the software development by Kaspar Arzner * discussion criticizing the complexity of Arzner's approach and its questionable relevance to observations with good statistics (e.g., can just use G1) * proposal for visibility-function approach (Hurford's plan A). Or (plan B) obtain a time series of individual modulations without gaps. Somebody needs to work on the software... Markus?! * Trottet asks about attenuation-corrected light curves. "You cannot remove the effect of attenuation on coarse energy bands" - Schwartz. * The quicklook data show "corrected" lightcurves. These are often quite misleading - "dangerous" even (Dennis). Examples of problems include the second orbit of April 21, 2002 - see the quicklook pages 4. Pileup (Smith) - deferred 5. Detector relative normalization (Smith) - deferred 6. RHESSI low-energy spectra (Dennis) * Can see Fe line complex even with thin and thick shutters in; peak energy and equivalent width can be determined quite well * April 21, 2002, showing different shutter regimes and pulse pileup on the Fe complex * Spectra, showing background Ge feature at 10 keV (includes Ge K edge). Also can rather clearly see the Fe-Ni complex at 8 keV, but this appears different in different detectors. The attenuators on the different detectors may be different (Schwartz) - SPEX assumes a quadratic sum of Poisson and choosable systematic errors in calculating chi^2 (Schwartz) * Fe abundance sensitivity ("wholly hypothetical" - Hudson) * Relatively good match between GOES and RHESSI temperatures in A1 mode * SPEX power-law-plus-Gaussian fits to get centroid energy independent of atomic (Mewe) model * Thick shutter response function clearly visible below 10 keV, but one can still fit to the Fe feature consistently * Detector-resolved centroid energies show clear variation prior to the initial development (prior to the first shutter motion) * Equivalent-width measures look interesting and hint at non-thermal continuum at low energies * Fe-Ni 8 keV feature also can be tracked, as can the continuum slope * Mewe vs Chianti: great model differences, probably largely in the area of ionization equilibrium 7. RHESSI low-energy spectra (Hudson) * See previous talk and on-line material at http://sprg.ssl.berkeley.edu/~hhudson/presentations/rhessinessi * Upper limits on gain variation over the mission can be gotten from the Fe feature centroid energy 8. Pileup (Smith) * Nature of spectral dependence of pileup (on blackboard!) and motivation for software for pileup estimation * Limitations of pileup correction are dominated by hardware uncertainties rather than the current estimation algorithm * 40 keV is the danger zone * Modulation effect on pileup is nonlinear! * Iterative procedure may not converge properly in a non-linear situation (Schwartz) * G9 does appear to show more pileup than the others in July 23 (Kontar answer to Hurford question) * Extensive discussion of ways to estimate the modulation factor (variance, power spectrum, etc.) for input to a pileup code (Schwartz, Hurford, Kontar, Smith) 9. Pileup in the inversion context (Kontar) * 00:30:00 - 00:30:20 spectrum from July 23 (Piana et al inversion) shows a dip at about 55 keV in the electron spectrum. It is also seen in the Johns-Krull inversions. The dip corresponds to a 5% residual at 40-50 keV. This could well be a 10% error on the 50% pileup correction (Smith). 10. High-energy spectroscopy (Smith) * Photopeak area vs energy 11. May 29 gamma-ray event? (Schwartz) * In the front segment, not the rear, but standard gain corrections have not yet been run yet 12. Other topics (Hurford) * Detector photometric mismatches (10% RMS - Smith, from flare of 29-Oct-02, 04:35-04:45) not presently understood * Misalignment of rotation axis and grid centerline not yet corrected and is energy-dependent (possibly 10%?) * Rotation-averaged grid transmission; coarsest grids are worse but on order 10% +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ June 5 Schwartz, Mrozek, Kontar, Trottet, Krucker, Hurford, Gan, Aschwanden, Smith, Hudson, Dennis 1. Imaging (Hurford) * Introduction 2. Image methods survey (Aschwanden) * Paper on line at http://www.lmsal.com/~aschwand/reprints/2003_photo includes a comprehensive tradeoff analysis on a representative time interval in a particular flare, exercising all of the parameters such as FOV, pixel size, etc. * Sorry, lost all my notes on the discussion, which was very interesting. Imaging and spectroscopy are consistent to 10% or so, but the different algorithms have very different properties. According to Schwartz and Hurford, both MEM methods are "broken", but they work sometimes. MEM-Sato prefers large pixels; MEM-vis prefers short integrations. Pixons look as good as they can. Clean also; forward method also within its limitations We need more of this kind of study, including simulations at whatever level of verisimilitude that they can have. 3. Discussion of Bob Lin question: can the images be used to estimate the background for integrated-sun spectroscopy? - "Great idea, but..." (Hurford) - G8 and G9 better - Could give good estimates of statistical error - There are grave systematic errors that make this difficult "A year's worth of work" (Schwartz) - Systematic errors vs energy are most dangerous (Hudson) but could do a simple experiment (using time-series variance as a measure?) on a known case. - Penalty of SNR would be heavy, and worst in the worst region of the spectrum - Could one just measure the spectrum of the modulated flux? (Schwartz) 4. Alternative background idea (Hudson): use the rear segment counts at low energies as predictors of the front-segment backgrounds. This should work well except for big events but needs software development. June 5 p.m. 1. Gamma-ray spectral observations (Smith) * Observables - The problem is extremely broad in terms of interpretation - "Are friendly neutrons detectable?" (McKinnon). No (Lin), "odds are very slim." * Observations and constraints * July 23 lightcurves * Overall spectrum * De-excitation redshifts: all lines redshifted, correlation with 1/M for both shift and broadening * Accelerated ions: observables, based on Murphy simulations * Model for all lines taken together - For alpha ratio of 0.5 and fixed spectral index @(3.75?) - Forward isotropic at cos = 0.8, a 36-degree bias angle relative to 73-degree flare location * 4D chi-2 minimizations (still not the whole problem, ie have ignored relative strengths; this is a vast problem) * Time variability suggests Fe/Ne differencs. Ne unusually low early, suggesting low-E later * Positron annihilation 8.1+-1.1 keV width * Alpha-alpha? * Neutron-capture @ 800 counts/keV peak spectral fluence 2. Imaging (Hurford) * 700 keV and 2.2 images with 300-500 keV contours overlaid show 2.2 source to be displaced to the S of the hard X-rays * The only feature in the 2.2 map that can be discussed is its centroid 3. Overinterpretation I (Emslie-Miller) * Miller calculations of acceleration vs loop lengths suggests that larger loops make protons, and more compact ones make electrons * Discussion of flare morphology, sources of wave energy, time delays, etc 4. Overinterpretation II (all) * 10 H inductance, 10^18 amps: one cannot expect charge separation due to reconnection electric fields * If this discussion were being held in St Andrews, we'd be talking about magnetic reconnection 5. Interpretation III (McKinnon-Toner) * Source temperature effects, taking advantage of the ~2 MeV threshold of the Ne line cross-section: fluence ratio of 1.63/6.13 vs T_e * Ne/O ratio is assumed, but it's a scary business since the energetics result depends upon this. Note also the apparent time dependence of Ne/O gamma-rays from the July 23 event. 6. Interpretation IV (Gan) * deferred 7. Albedo I (C. Alexander) * Timing delay, spectroscopic signature, "halo objects" * Spectral bumps: 40-50 keV, disappearing for softer spectra * Discussion of whether inverse space is the right place to do the comparisons * Spectral bumps are not observed.. Discussion of remarkable coincidences in Compton and pileup physics. * Bethe-Heitler version in the works 8. Albedo II (Schmahl) * Large extent => overresolution, cf. work with Hurford on source sizes * Relative amplitudes via unpixelized Fourier fitting, back-projection, C-statistic * Compact round sources picked 12-25 keV * Albedo theory from isotropic assumption => Moffat functions; see Brown et al., 1974, based on isotropic downward component. The albedo brightness is "up for grabs", ie not found in Bai-Ramaty * Relative brightnesses => "basis functions" => Fourier transforms * Fits to observed relative visibility amplitudes => actual numbers, but these are _not_ to believed. * Energy dependence, ellipticity, tilt of albedo patch, phase shift, anisotropic scattering, extended primary source * Observed spectral dependence of visibility spectra is consistent with the albedo explanation. * 50% contour ~ 3.5 x height, 90% 9.9 x * Fourier transforms of albedo patches * 2002 July 3 event, find G9 shows map shifts with energy - albedo? * Full characterization, first step: 6 parameter fit, including albedo amplitude, height, xy, amplitude, width (Bai-Ramaty homework would reduce this to 5 parameters) * Do it all successfully and also one might determine the primary directivity! 9. Albedo III (St.-Hilaire) * Kontar and Alexander papers applied to energy budget considerations * "cutoff" spectral model, zero below E_k (not too physical?) * "kink" model, flat below E_k * E_k is most important factor, but not easy to observe (eg pileup; example of 2002/5/27 event) * Intersection of thermal and non-thermal components => UL on E_k, lower limit on power * ionization and albedo corrections increase the photon flux, leading to an overestimation of P_kin * IDL routine 1 to model range, gamma, E_k dependences * IDL routine 2 => P_kin range * April 17, 2002 => E_kin/E_th ~ 15-30 in kink model, but smaller in cutoff model (5-10 x) * Feb. 26, 2002: factors 10-90, 6-30 in the two models * But everything is OK if one moves E_k up to the break energy * Filling factor probably does not help reconcile; smaller volumes mean smaller thermal energies. Hudson suggestions of ff>1 rejected. 6. Interpretation (Gan) * Information from temporal evolution of spectrum, example of Dec. 16, 1998 (Gan-Rieger 1999) * Background-subtracted spectra from July 23 and model: single power law > 300 keV, 2.223, 5 broad lines with fixed energies. Big difference with Share time series at 511. * Simplified model may not work, because it lacks the detector response in detail? Smith comment - shouldn't matter too much, but worry about use of full response matrix. * Modeling the 3 bands (.511, 2.23, 4-7) but just using Share et al results: suggest softening spectral index * Inconclusive results; we need a bigger flare * What about pion contribution, ie power law extending above 500 MeV? suggest different episodes in which this could help with the evolution * Worry about actual width of 2.223 MeV * Artefact at 3 MeV is not a line 10. Imaging spectroscopy discussion (Hurford et al.) Gordon's table shows source, symptoms, significant if, affects chi2, to minimize, fixable, these are the various sources of systematic errors that we worry about. Please note that this is preliminary and will be made more official as Gordon develops it further. - Statistics - Missing uv converage - Inadequate resolution - Inadequate coarse grid resolution - Finite time bins - Grid response errors - Attenuator response errors - Detector responses errors - Dead time calculation - Data gap correction - Correlation between background and pointing - Flat-fielding close to rotation axis - Energy calibration - Neglect of off-diagonal terms - Wide energy bands (constancy of response over band) - Large FOV (constancy of response over FOV) - Flux variations - Morphology variations with time and/or energy - Aspect errors (limiting factor for G1) - Neglect of harmonics - Code bugs - Operational errors - Bipolar pulses (several hundred keV in rear segment needed) - K escape from GeD (slightly redundant here, but most important) - Scattering in lower grids - Diffraction in grids - Excessive mod_pat_skip - Correlation between grid bridges - Use of incorrect spin axis offset (a biggie) - Biased modulation error - Background subtraction - Inappropriate weighting - Backprojection mirror sources and sidelobes - Backprojection close to axis - CLEANing of extended sources - MEM-VIS and variable sources - MEM pixel size - Inappropriate algorithm parameters - Forward-fitting an inappropriate model - Forward-fit background subtraction - Pixon background - Forward-fitting subcoll selection - Polar-Cartesian sampling - Live-time response mismatch - D8 transmitter issue - Ignoring precipitation events - Inappropriate goodness-of-fit criterion - SPEX energy-edge definitions - Inappropriate use of theoretical models - Image dynamic range limitation - Detector photometric calibration - User gullibility ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ June 6 (joint discussion with group 2) Emslie, Johns-Krull, Gan, Schmahl, Vilmer, Lin, Smith, Schweartz, Trottet, Krucker, Mrozek, Kontar, Hudson, Aschwanden, St.-Hilaire, Dennis, Conway, C. Alexander, Veronig, Hurford 1. Imaging Spectroscopy (Krucker) * CLEAN imaging of July 23, with forward fit to help identify "sources" * Error estimation. CLEAN residual map used to estimate noise level from outside the image area: maximum of noise map taken as "five sigma" (note that fluctuation is roughly consistent). This approach may overestimate the errors systematically. Hurford notes that _including_ the real source components leads to a kind of upper limit on the uncertainties. * An example @00:29 UT for footpoints f2 and f3, using grids 3-8 <20 keV and 2-8 above. Lose sight above 50 ke. * Discussion of pileup effects. Hurford ingenious proposal to re-order in terms of dead time, then to apply a smoothly varying pileup correction, then to restore to time-series order and proceed with imaging. * Pileup illustration from April 21, 2002, a wonderful example of many systematic errors! Maximum of pileup/true ratio peaks at about 50 keV. At thick-shutter movement, the pileup correction becomes negligible everywhere. * More discussion of pileup. Hurford worry about 2-D sorting for spectral variability; ingenious Emslie remark that the peak region of the pulse-height spectrum is all that really matters, so maybe a 1-D resorting would be pretty good. * April 21 example, comparing A1 and A3 states shows the spurious appearance of a hard coronal source. 2. Solicitation of general discussion * The status of the "stacker" (Schwartz): at each time, rebin into phase of map center and rotation angles (counts, live time etc), and then use in succeeding software. Now struggling in software with some optimization. Advantage is that one can sum over short time bins and go directly to (u,v) visibilities. * Aschwanden idea for a new imaging method: "ribbon-constraint imaging" as an extension of forward-fitting, ie to take a template from some other wavelength and then proceed. Could do this also with CLEAN by constraining the image locations. 3. Smith: let us make priorities of the things we have to do starting from spectroscopy to imaging spectroscopy * Spectroscopy: it looks like it is pile-up correction (Smith) - Schwartz: we need to make the script to build to do automatic corrections - Dennis: Is it going to be compatible with new object oriented version of spex ? - Schwartz: yes - Smith : Before pile-up correction we need correct for decimation - Schwartz: I can do existing script for decimation correction to be more robust * Imaging: It will be really important to have low energy part of the footpoint spectra (lin) - hurford: indeed we have to improve dynamic range .... 4. McKinnon: Ne/O line ratio considerations 5. Hudson: 2.2 MeV survey * See http://sprg.ssl.berkeley.edu/~hhudson/presentations/rhessinessi/gr_stats.*