Anomalies in low-energy Gamma-Ray Burst spectra with the Fermi Gamma-Ray Burst Monitor

1.4 MB·English

by  Dave Tierney

Astronomy&Astrophysicsmanuscriptno.LEE˙AA˙v10 (cid:13)c ESO2013
January22,2013
Anomalies in low-energy Gamma-Ray Burst spectra with the Fermi
Gamma-Ray Burst Monitor
D.Tierney1,S.McBreen1,R.D.Preece2,G.Fitzpatrick1,S.Foley1,S.Guiriec3,E.Bissaldi4,M.S.Briggs2,
J.M.Burgess2,V.Connaughton2,A.Goldstein2,J.Greiner5,D.Gruber5,C.Kouveliotou6,S.McGlynn5,7,
W.S.Paciesas8,V.Pelassa2 andA.vonKienlin5
1 UniversityCollegeDublin,Belfield,Dublin4,Dublin,Ireland
e-mail:[email protected]
2 PhysicsDepartment,UniversityofAlabamainHuntsville,320SparkmanDrive,Huntsville,AL35805,USA
3 NASAGoddardSpaceFlightCenter,Greenbelt,MD20771,USA
3 4 InstituteforAstroandParticlePhysics,UniversityofInnsbruck,Technikerstrasse25,6020Innsbruck,Austria
1 5 Max-Planck-Institutfu¨rextraterrestrischePhysik,Giessenbachstrasse1,85748Garching,Germany
0 6 SpaceScienceOffice,VP62,NASA/MarshallSpaceFlightCenter,Huntsville,AL35812,USA
2 7 Exzellenz-ClusterUniverse,TechnischeUniversita¨tMu¨nchen,Boltzmannstrasse2,85748Garching,Germany
8 ScienceandTechnologyInstitute,UniversitiesSpaceResearchAssociation,320SparkmanDrive,Huntsville,AL35805,USA
n
a Preprintonlineversion:January22,2013
J

1 ABSTRACT
2
Context. A Band function has become the standard spectral function used to describe the prompt emission spectra of gamma-ray

] bursts(GRBs).However,deviationsfromthisfunctionhavepreviouslybeenobservedinGRBsdetectedbyBATSEandinindividual
E
GRBsfromtheFermiera.
H Aims.WepresentasystematicandrigoroussearchforspectraldeviationsfromaBandfunctionatlowenergiesinasampleofthe
firsttwoyearsofhighfluence,longburstsdetectedbytheFermiGamma-RayBurstMonitor(GBM).Thesamplecontains45bursts
.
h withafluencegreaterthan2×10−5erg/cm2(10-1000keV).
p Methods. An extrapolated fit method is used to search for low-energy spectral anomalies, whereby a Band function is fit above
- a variable low-energy threshold and then the best fit function is extrapolated to lower energy data. Deviations are quantified by
o
examining residuals derived from the extrapolated function and the data and their significance is determined via comprehensive
r
t simulationswhichaccountfortheinstrumentresponse.Thismethodwasemployedforbothtime-integratedburstspectraandtime-
s resolvedbinsdefinedbyasignaltonoiseratioof25σand50σ.
a
Results.Significantdeviationsareevidentin3bursts(GRB081215A,GRB090424andGRB090902B)inthetime-integratedsample
[
(∼7%)and5bursts(GRB090323,GRB090424,GRB090820,GRB090902BandGRB090926A)inthetime-resolvedsample(∼

1 11%).
v Conclusions. The advantage of the systematic, blind search analysis is that it can demonstrate the requirement for an additional
9 spectralcomponentwithoutanypriorknowledgeofthenatureofthatextracomponent.Deviationsarefoundinalargefractionof
5 highfluenceGRBs;fainterGRBsmaynothavesufficientstatisticsfordeviationstobefoundusingthismethod.
8
Keywords.Gamma-rayburst:general–Methods:dataanalysis–Techniques:spectroscopic
4
.
1
0
1. Introduction The lightcurves of GRBs are highly variable and a number of
3
studiesintotheirtemporalpropertieshavebeenperformed(e.g.,
1
: Gamma-raybursts(GRBs)arethemostluminouseventsinthe Quilliganetal.2002;Hakkila&Preece2011;Bhatetal.2012).
v OurworkassumesthataBandfunction(Bandetal.1993)is
i universe and can briefly be summarised as having high-energy
X the best fit function forGRB spectra in the GBM energy range
prompt emission followed by a multi-wavelength fading after-
andattemptstoquantifythenumberofburststhatdeviatefrom
r glow (e.g., Vedrenne & Atteia 2009; Kann et al. 2011). The
a isotropic energy produced by a typical GRB is E ∼ 1051 erg thisfunction.Ingeneral,aBandfunctioncanbeconstrainedbet-
iso teratlowerenergiescomparedtohigherenergiesasmorecounts
(e.g., Frail et al. 2001) and in some cases reaches up to 1054
areobservedatlowerenergies.Thisprovidesstrictlimitsonthe
erg(e.g.,Greineretal.2009;McBreenetal.2010;Cenkoetal.
αparameterandmakesitpossibletosearchfordeviationstothe
2011). The prompt emission of GRBs has been detected over
fitfunction.TheBandmodelisdefinedas
a wide spectral range from keV to GeV energies and is gener-
as2emlt0lay0ol9.owb2th)e0,ll1yla1mb;qroRuodayksedeil-enltehedpetorbawmly.ea2orl0-nlc1eao1wom;rA(peaox.gnce.eolsnmBstoab(nniend.egatet.ai,tolP.anr2l.eo0e1f1c92teh9),e32a;0fn0oA0leblx;odtwGroaiunenitgroi:aenlac-. N(E)=AA(cid:16)(cid:16)1(1αE00−00β(cid:17)()2αE+epαexa)kp(cid:17)α(cid:18)−−βEeE(x2p+epaαk()β(cid:19)−α)(cid:16)1E00(cid:17)β iiff EE ≥< EEcc (1)
thermal power-law component extending to high energies (e.g.
where
Gonza´lez et al. 2003; Kaneko et al. 2008; Abdo et al. 2009a) E
or a cut-off in the MeV regime (e.g., Ackermann et al. 2012b). Ec =(α−β)2+peaαk (2)
1
D.Tierneyetal.:Anomaliesinlow-energyGRBspectrawithFermiGBM
and A is the amplitude in photons s−1 cm−2 keV−1, Table 1. Comparison of properties of the GBM NaIs with the
α is the low-energy power-law index, E is the ν F BATSESDs.
peak ν
peak energy in keV and β is the high-energy power-law index.
BATSESD1 GBMNaI2
Material NaI NaI
TheemissionmechanismsforGRBshavegenerallybeende-
Number 8 12
scribed as thermal or non-thermal with the thermal emission
Area 126cm2 126cm2
characterised by the radiation emitted from a cooling plasma
Thickness 7.62cm 1.27cm
(e.g.Me´sza´ros&Rees2000).However,thermalemissionneed EnergyRange 30keV-10MeV∗ 8keV-1MeV
nottaketheformofastandardPlanckianmodelbutmaytakea
SpectralBinning
more complex form depending on several factors including the 8-20keV 1bin ∼12
relative leptonic and hadronic populations, viewing angles and
sourceregion(e.g.,Rydeetal.2011).Non-thermalemissioncan
take the form of synchrotron emission from the extreme mag-
Notes.(*)Thisenergyrangevarieddependingonthegainofthedetec-
neticfieldsneededtocreateaGRB(e.g.Me´sza´ros2002).Strong
torswherethelowerendoftherangecouldbeaslowas5keVfora
magnetic fields may also break and reconnect releasing energy
highgainsetting(Preeceetal.1996).
in the form of gamma-ray photons (e.g. Zhang & Yan 2011).
AlthoughaBandfunctionisnotaphysicalmodel,itsuccessfully
References.(1)Preeceetal.(1996),(2)Meeganetal.(2009)
models a large number of GRBs and it is useful to investigate
thenumberofeventswhichrequireadditionalparamaters.Ifan
additionalblackbodyisobserved,thetemperatureofthephoto-
spherecanbededuced(Rees&Me´sza´ros2005)whereasanun- pseudo-logarithmically across the energy ranges of each detec-
derlyingpower-lawcanprovideconstraintsonemissionmecha- tor. The lowest and highest energy channels in each detector
nisms (e.g. Asano et al. 2009; Ryde et al. 2010) and the Extra- are generally ignored in the analysis because of uncertainties
galacticBackgroundLight(EBL)(Ackermannetal.2011).An in the instrument response. Extensive ground calibration was
overview of the observational and physical interpretations of carried out on the detectors pre-launch (Bissaldi et al. 2009)
theresultsfromtheFermieraarepresentedbyBhat&Guiriec andin flightby comparisonto otherinstruments (INTEGRAL-
(2011). ISGRI:Tierneyetal.(2011),INTEGRAL-SPI:vonKienlinetal.
A number of papers discuss interesting bursts with unusual (2009)andSwift-BAT:Stamatikos(2009)).Thesecalibrationre-
spectral behaviour in the keV range (e.g., Guiriec et al. 2011; sults are consistent and do not show any major changes from
Ryde et al. 2011; Axelsson et al. 2012). It is important to note the on-ground calibration. Additional calibration work has also
thatthereisabiasintheliteraturewherebyinterestingburstsare been carried out using the Crab Nebula with Earth occultation
published more than bursts that conform to the standard GRB techniques(Wilson-Hodgeetal.2012),nuclearlinesfromsolar
models.Anotherbiasthatcanskewtheapparentnumberofin- flares (Ackermann et al. 2012a), positron/electron annihilation
terestingburstsisthatonlyburststhatlookatypicalinitiallyare lines(Briggsetal.2011)andspectralanalysisofSoftGamma-
investigatedfurther.Arigorousinvestigationmustbecarriedout rayRepeaters(SGRs)(Linetal.2012).
onasampleofGRBsinasystematicwaytoobtainthetruefrac-
tion of GRBs with observable additional features. Any investi-
1.2. FermiGBMandBATSE/CGROComparison
gation must also be performed blindly to decrease the risk of
interval selection effects which could bias the study. Here we Acomprehensivestudyexaminingdeviationsatthelowerendof
presentasystematicsearchfordeviationsfromaBandfunction thespectrumhasnotbeencarriedoutsincePreeceetal.(1996)
inthehighestfluenceburstsinthefirstcatalogsfromtheFermi investigatedtime-integratedGRBspectrausingthespectroscopy
Gamma-rayBurstMonitor(Paciesasetal.2012;Goldsteinetal. detectors(SDs)ontheBurstAndTransientSourceExperiment
2012). (BATSE) on board the Compton Gamma Ray Observatory
(CGRO)(Gehrelsetal.1992).Theresultsofthisstudyshowed
that ∼ 14% of a sample of 86 GRBs contained significant low-
1.1. FermiGamma-rayBurstMonitor
energy excesses; no significant deficits were observed in the
TheFermiGamma-raySpaceTelescope,launchedon2008June sample.AcomparisonbetweentheeffectiveareasofBATSEand
11, has an energy range spanning several decades (∼8 keV to GBMisgiveninFigure1.TheGBMNaIdetectorsareofsimilar
∼ 300 GeV) and is ideal to explore the low-energy regime specifications to the BATSE SDs (see Table 1). As the BATSE
of GRBs. Fermi consists of two instruments, the Large Area SD detectors were significantly thicker than the GBM NaI de-
Telescope (LAT) operating between ∼ 20 MeV to ∼ 300 GeV tectors,theBATSESDscoulddetecthigherenergyphotonsthan
(Atwoodetal.2009)andtheGamma-RayBurstMonitor(GBM) the GBM NaIs. This is compensated for on GBM by using the
operatingbetween8keV-40MeV(Meeganetal.2009). BGOdetectorsforhigh-energyconstraints.
GBM consists of two types of detectors - twelve Sodium The relative effective areas for each detector are presented
Iodide (NaI) scintillating crystals operating between 8 - 1000 in Figure 1 showing that in the crucial energy range of interest
keV and two Bismuth Germanate (BGO) scintillating crystals (below∼30keV),theNaIshaveagreatereffectiveareathanthe
operating between 0.15 - 40 MeV. The NaI detectors are ar- SDdetectors.Additionally,whilePreeceetal.(1996)onlyused
rangedinclustersofthreearoundtheedgesofthesatelliteand a single SD detector, multiple NaI detectors were used in our
theBGOsarelocatedonopposingsidesofthesatellitealigned analysisinallbut3GRBs(see§2).GBMalsohastheadvantage
perpendiculartotheLATboresight.Asthedetectorshavenoac- ofgreatereffectiveareaathigherenergiesusingtheBGOs.This
tiveshieldandareuncollimated,GBMobservestheentireunoc- providesbetterconstraintsonβand E whichhelpconstrain
peak
cultedsky.FurtherdetailscanbefoundinMeeganetal.(2009). theoverallspectralmodelappliedtothedata.
BothNaIandBGOdetectorscollectcountsin4096channels ToprobethelowerenergiesusingtheBATSESDs,twodata
compressed into 128 energy channels, with boundaries spaced types were required. The first data type was the Spectroscopy
2
D.Tierneyetal.:Anomaliesinlow-energyGRBspectrawithFermiGBM
ormoreNaIswereusedfor42outofthe45GRBs.Thebright-
estBGOdetectorwasusedinallcasestoconstrainthespectral
modelathighenergies.
As the signal to noise ratio (S/N) in these bursts is quite
high, the influence of background fluctuations has a less statis-
ticallysignificanteffectduringthespectralfittingprocesscom-
pared to weaker bursts. Individual detectors were checked for
blockages whereby the source photons passed through part of
thespacecraftbeforeenteringadetector.Thiscancausephotons
to be scattered to different energies and lower energy photons
to be absorbed beyond the ability of the detector response ma-
trices(DRMs)toaccuratelyreconstructthephotonsincidenton
the entire spacecraft. An automated tool was initially used to
checkforlineofsightblockagesbetweentheGRBandadetec-
tor.Furthermanualanalysiswasperformedtoremoveanyaddi-
tionalblockeddetectors.Detectorsthatareblockedhaveanot-
Fig.1.PlotcomparingtheeffectiveareasofGBMvBATSEper
icable deficiency of low-energy counts which is not consistent
detector.Thetotalareaexposedtoaparticularskydirectionde-
with other detectors that have similar source angles. Detectors
pends on the orientation of each detector and number of detec-
whichwerenotblockedandhadanacceptablesourceanglewere
torsexposed.AtlowenergiestheGBMNaIdetectorshavemore
definedas‘good’andusedinthesubsequentanalysis.
effectiveareathantheBATSESDsandathighenergiestheGBM
Figure 2 shows the spectral properties of this sub-sample
BGOdetectorshavemoreeffectiveareathattheBATSELADs
(α and E ) relative to the overall GBM spectral catalog
(Meeganetal.2009).Thedarklinesrepresentthephotopeakef- peak
(Goldstein et al. 2012). The distributions show that the sub-
fective area where the detected energy is the same (within the
sampleselectedheredisplayssimilarcharacteristicstotheentire
energy resolution) as the incident energy and the lighter lines
GBMspectralcatalogue.
represent the total effective area which includes the photopeak
plusthecaseswheretheinstrumentdetectsonlypartoftheinci-
dentphotonenergy.(Acolorversionofthisfigureisavailablein
theonlinejournal.)
3. Method
Thesamplewasanalysedusingtwomethods.Thefirstcompares
Discriminatordatachannels(DISCSP)whichcontainedfourin- thedatatotheextrapolatedfitofthefunctioninthelow-energy
tegral channels over the entire energy range ∼ 5 keV - 2 MeV regimeandthesecondcomparesthechangeinthespectralindex
(forthehighestgainsetting).Theupperedgeofthelowestdata α when the energy range of the data is shortened. These two
channel (DISCSP 1) was set to the Lower-Level Discriminator methods are used in combination to identify deviations from a
(LLD) threshold of the Spectroscopy High-Energy Resolution simplepower-lawatlowerenergies.
Burst(SHERB)data(∼10keVforthehighestgainsetting).The
For an additional component to be significantly detected in
SHERB data comprised 256 spectral energy channels between
the spectrum of a burst, it must either be continuously present
theLLDand∼2MeV.ByusingajointfitbetweentheSHERB
throughout the entire burst or very strongly present in certain
data and the single DISCSP1 data point, Preece et al. (1996)
sections of the burst. An additional component may be present
determined the deviation of this single data point to the model.
in certain sections of a GRB but not be intense enough to sig-
GBMhastheadvantageofasinglestandarddatatypefromlow
nificantly alter the overall spectrum of the burst. Therefore the
tohighenergies.Thisreducesthechanceofanyincongruitybe-
spectralfittingwasperformedonthespectrumoftheentireburst
tweendatasets.Thedatatypealsohasbetterspectralresolution
(time-integrated) and on significant time slices (time-resolved)
intheregionofinterest(seeTable1)sotrendsinthedatacanbe
ofthedata.AllspectralfittingwasperformedusingtheRMFIT
more easily distinguished. GBM also has the advantage of bet- softwarepackage1andminormodificationsweremadetoauto-
tersourcelocationsprovidedbyotherinstrumentssuchasSwift
matethefittingprocess.
andtheLATwhichimprovestheabilitytoaccuratelymodelthe
instrumentresponse.OutofthesampleofGRBsanalysedinthis
work,47%haveasub-degreelocalisation. 3.1. Time-IntegratedSingleFit
CSPECdatawereusedforeach‘good’detectoroverthefullen-
ergyrangeofGBM(8keV-40MeV).Thesedatahaveanenergy
2. SampleSelection
resolutionof128channelsandatemporalresolutionof4spre-
Thesamplewasdrawnfromthefirst2yearsoftriggeredGBM triggerchangingto1spost-triggeruntilT0+600s.Usingthe
GRBs(14th July2008-13th July2010).Burstsaboveafluence lightcurve,atimeregionencompassingthemainemissionphase
of 2 × 10−5 erg cm−2 (10 - 1000 keV) were selected so that oftheGRBwasselectedbyeye.Apolynomialbackgroundwas
the spectral parameters could be constrained in the fitting pro- thenfittoeachGRBlightcurvebyselectingbackgroundregions
cess(seehowever§3.5).Thisgaveasampleof45bursts,which before and after the prompt phase. Multiple response matrices
formedthebrightest9%ofthe491GRBsinthefirstGBMGRB were used in the fitting process to account for spacecraft slew-
catalog(Paciesasetal.2012).Inaddition,36outofthe45GRBs ingrelativetothesource.Inthisprocessanewresponseismade
(80%)havesignificantemissioninatleastoneBGOdetectoras whenthespacecraftslewsbymorethan2degrees,whichcanbe
definedbyBissaldietal.(2011).OnlyNaIdetectorswithsource importantinlongbursts,andisespeciallyimportantifthespace-
angleslessthan60◦wereusedinthespectralanalysistolimitthe
effectoftheuncertaintiesintheoff-axisdetectorresponse.Two 1 http://fermi.gsfc.nasa.gov/ssc/data/analysis/user/
3
D.Tierneyetal.:Anomaliesinlow-energyGRBspectrawithFermiGBM
140 120
Catalog α Catalog Epeak
120 Sample α Sample Epeak
100
100
80
80
60
60
40
40
20
20
0 0
2.0 1.5 1.0 0.5 0.0 0.5 1.0 101 102 103
α Epeak (keV)
(a) (b)
Fig.2.Histogramscomparinga)αandb)E inthehighfluencesamplefromourworktoasampleobtainedfromGoldsteinetal.
peak
(2012).
craftexecutesanautonomousrepoint.Asimilarmethodwasalso range should be consistent with a fit performed over the entire
employedbyGoldsteinetal.(2012). energyrangeintheabsenceofanyspectralexcessesordeficits
Aninitialtime-integratedspectralfitwasperformedusinga withrespecttothefunction.Theparametersobtainedbyfitting
BandfunctionforeachGRBinthesample.Fitresidualswerede- from8keV-40MeVshouldbeconsistentwiththoseobtained
terminedbythenumberofstandarddeviationsthatseparatethe by fitting from 15 keV - 40 MeV when no additional compo-
datafromaBandfunction.Theresidualsweresummedbetween nents are present. The difference between the single fit method
8 keV and a variable Low-Energy Threshold (LET) in order to andextrapolatedfitmethodisshowninFigure3.MinimizingC-
search for excesses or deficits. The LET was set at 15, 20, 25, statwilltendtoproduceafitthatbalancesexcessesanddeficits
30, 50 and 100 keV. However, this technique is not optimal as regardless of the physical origin of the deviation from a Band
the fitting algorithm forces the function through any deviations function. In the single fit method, the deviation at low energies
that are present and minimises any deviations in any particular results in strong fluctuations throughout the entire spectral en-
partofthespectrum.Toreducethisissue,theLETwasusedas ergyrange.Whentheextrapolatedfitmethodisused,thespec-
alowerenergyboundforspectralfittingandthedatapointsbe- tral deviations are observed more clearly at lower energies and
low the LET were compared to an extrapolated version of the the effects on the remaining spectrum are reduced. Features in
function.Thisextrapolatedfitisdefinedinthenextsection. certainenergybandsarenotphysicallycomparablebetweendif-
Although the sample was defined to be composed of high ferent GRBs because the energy bands are defined in the ob-
fluenceGRBs,thisdoesnotnecessarilyguaranteealargenum- serverframeandredshiftsarenotknownforallGRBs.However,
ber of counts in all energy channels. Due to the potential low whenperformingananalysisonthemanifestationofobservable
countrate/binratio,thefit(minimization)wasperformedusing deviationsincertainenergybands,fitsusingthesameLETcan
Castor statistics. This is similar to Cash-statistics (Cash 1979). becomparedusingthismethod.
CastorstatisticsassumePoissonuncertaintiesperbincompared Thelow-energypower-lawindex,α,mustbereasonablywell
to χ2 statistics which assumes Gaussian uncertainties per bin. constrained in orderto confidently calculate the significance of
Theseareequivalentinthehighcountregimebutdivergeinthe anydeviations.ThisimpliesthatE andtheLETmustbesuf-
peak
low count regime (less than ∼ 10 counts / bin). A similar min- ficiently different to ensure that α can be constrained over the
imisationmethodwasusedbyGoldsteinetal.(2012). shortenedenergyrange.ThetechniquealsorequiresthatE is
peak
greaterthantheLET.
ForeachGRB,thespectralresidualsbetween8keVandthe
3.2. Time-IntegratedExtrapolatedFit
LET are summed in an individual detector. The overall value
An improved extrapolated fitting technique was devised such foraparticularGRBisthearithmeticmeanfordeviationsinall
that a Band function was fit in the energy range from the vari- detectors used in the analysis of the burst. The Iodine K-edge
ableLETto∼40MeV.Thefunctionobtainedfromapplyingthe region of the spectrum around 33 keV can lead to larger resid-
modelinthisnarrowerrangewasthenextrapolatedtothelower ualscausedbysystematicsratherthanaprocessintrinsictothe
energiesinordertoascertainhowwellthefunctiondescribedthe source. However, a strong K-edge effect was only noted in ∼
databelowtheLET.Deviationspresentbetween∼8keVandthe 5 GRBs and dominated only 2 spectral bins. If the K-edge is
LETarequantifiedbysummingtheresidualsbetween∼8keV present between 8 keV and the LET, it only weakly affects the
andtheLET. summing of residuals over this range. In reality it is only rele-
Oneoftheassumptionsoftheextrapolatedfittingtechnique vantwhenLET=50or100keVandtheeffectwasnotdeemed
is that the function parameters obtained over a shorter energy significantenoughtoalterthemethod.
4
D.Tierneyetal.:Anomaliesinlow-energyGRBspectrawithFermiGBM
(a) (b)
Fig.3.Comparingthesingletime-integratedfittotheextrapolatedfittingtechnique.a)SingleBandfunctionfittoGRB090902B
from8keV-40MeV(α=-0.99+0.01,E =996.8+14.1 ,β=-5.44+0.89 ).b)BandfunctionfittoGRB090902Bfrom30keV-40
−0.01 peak −14.2 −53.50
MeVandextrapolateddownto8keV(α=-0.85+0.01,E =837.4+11.9,β=-4.42+0.38).
−0.01 peak −11.6 −0.71
LETwerethennormalisedbycalculatingthenumberofstandard
deviationsbywhichthedatavariedfromthemeanofthesimu-
lateddistribution.IfthedataforaGRBdiffersignificantlyfrom
3.3. Time-ResolvedExtrapolatedFit
its simulated distribution, then it is claimed that a low-energy
The spectrum of a GRB can evolve over the prompt emission deviationispresent.
interval(e.g.Preeceetal.2000)andatime-resolvedanalysisis A representation of the overall distribution was simulated
required to account for such evolution. An S/N approach was to provide an initial insight into the expected distribution of
employed over a basic time-resolved analysis (i.e. splitting the residuals assuming that all spectra were correctly described by
bursts into time sections) to ensure significant counts per bin. a Band function. Five GRBs were selected from the sample
The selected region for each GRB used in the time-integrated which represented a broad range of peak energies, with E
peak
fitting (full burst) was binned to the 25σ and 50σ level above = 161, 276, 444, 484 and 1010 keV. These GRBs were simu-
background in the brightest detector. These intervals were then lated and the resulting distribution of residuals in the absence
usedtodefinebinsforallotherdetectorsinaparticularGRB.A of excesses/deficits are presented in Figure 4. Individual simu-
fitwasthenperformedoneachnewlydefinedbin. lations were then performed on a range of time-integrated and
time-resolvedsectionsofGRBs.
3.4. Simulations
ThegoodnessoffitofthedatabelowtheLETandthefunction
fitathigherenergiescannoteasilybeevaluatedusingC-Statas 3.5. DataCuts
thereisnoanalyticresultavailabletoconvertfromthefitstatistic
toameasureofgoodnessoffit.Thereforeitisdifficulttoquan- Several cuts were applied to the data before analysis. If any fit
tifythesignificanceofadeviationorevenanacceptablerangeof failed in the automated fitting process due to a lack of spectral
valueswhichoccurintheabsenceofanexcess/deficitatlowen- constraints,itwasexcludedfromthesample.Arangeofinitial
ergies. Simulations are therefore necessary to quantify whether parameterswerepassedtothefittingprocessinordertoobtaina
adeviationisconsistentwiththeextrapolatedfit. satisfactorysolution,but9fitsfailedinthetime-resolvedfitting
Aboot-strappingmethodwasusedtocomparetheactualre- processdespitealternativeinitialparametersandwereexcluded
sults from the data with simulated results. In order to perform fromthesample.Thesefitswereindividuallyexaminedandthe
the simulations, the background and source must be simulated. failurewasusuallyduetolow Epeak andalackofcountsabove
The background was simulated using a similar level to that in Epeak.Nofitsfailedinthetime-integratedanalysis.
the real data. To obtain the source region, a perfect Band func- Asthelow-energyspectralindexαonlyapproachesapower-
tion specific to each GRB, or time interval selected, was simu- law in the asymptotic limit, a sufficient data range is needed to
lated.Multiplecountdistributionswerecreatedusingthisfunc- obtain reasonable constraints on the index. If E is too close
peak
tion(varyingbyPoissonnoise).Eachsetofparameterswassim- totheLET,twoissuesmayarise1)αwillnotapproachthelimit
ulated∼1000times. and2)therewillnotbesufficientdatatoconstrainαresultingin
For each simulated spectrum, a distribution of low-energy largefluctuations(e.g.LET=50keV, E =60keV).Awell
peak
deviationsusingdifferentLETscanbecompiled.Thesimulated constrainedαparameterisrequiredbecauseslightvariationsin
distributionswerethenfitwithaGaussiandistributiontoobtain αcancauselargespectraldeviationswhenextrapolatedtolower
avalueforthemeanandstandarddeviation(seeFigure5).The energies.Foraspectrumtobeacceptedinthesamplethefollow-
data obtained by summing the low-energy residuals below the ingcriteriamustbemet:
5
D.Tierneyetal.:Anomaliesinlow-energyGRBspectrawithFermiGBM
800 9
LET = 25 keV LET = 25 keV
700 8
7
600
6
500
5
400
4
300
3
200
2
100 1
0 0
20 15 10 5 0 5 10 15 20 40 20 0 20 40 60 80 100 120
Summed Sigma Residuals Summed Sigma Residuals
(a) (b)
Fig.4. Comparing the time-integrated data to a simulated distribution. a) Combined distribution of low-energy deviations of 5
individualGRBsassumingaperfectBandfunction(nodeviationspresent).TheextrapolatedfitmethodwasusedwithLET=25
keV. b) The low-energy residuals of the GRBs in the sample that survived the data cuts (see § 3.5) for LET = 25 keV. The data
distributionb)isbroadlysimilartothesimulateddistributiona).Thedatapointoutsidethemaindatadistribution(ontheright)is
GRB090902B.
Table2.NumberofsampleGRBs/GRBintervalspre-andpost- GRBwasinvestigatedfurther.Usuallyaburstwithastrongde-
cuts, where TI are the time-integrated results and 25σ/50σ are viation in α would vary across multiple different LETs. If an
thetime-resolvedresults. interval had a variation in α and there was a deviation present
inthelow-energyresiduals,thisintervalwasconcludedtohave
LET TI TI 25σ 25σ 50σ 50σ asignificantspectraldeviationfromaBandfunction.AllGRBs
Pre Post Pre Post Pre Post
that had large deviations in the low-energy residuals displayed
15 45 42 672 457 309 230
theexpectedvariationsinα.
20 45 41 672 438 309 227
25 45 41 672 389 309 222
30 45 41 672 338 309 217 3.7. SummaryofMethod
50 45 23 672 189 309 117
100 45 22 672 81 309 83 Abriefsummaryofthemethodfortime-resolvedanalysisispre-
sentedbelow.
– ForLET=15,20,25,30keV,E >100keVisrequired 1. The highest fluence GRBs from the first 2 years were se-
peak
– ForLETs=50or100keV,E >200keVisrequired lectedforanalysis.
– All intervals with a large pproeapkortional error on E , ∆ 2. Foreachburst,NaIdetectors<60◦tothesourceandwithout
peak
E /E >0.45wereexcluded. blockages were selected. Multiple NaIs and a single BGO
peak peak
– Allintervalswithanerroronα>0.2werediscardedtoen- wereselectedinmostcases.
surethatαcanbereasonablyextrapolatedbeyondtherange 3. Backgroundandsourceregionswereselectedforanindivid-
ofthefit. ualburst.
4. The main emission (time-integrated) interval of the GRB
ThefinalsampleafterthecutscontainsGRBsandintervalswith wasselectedbyeye.
good statistics and well constrained spectral parameters. Table 5. TheGRBswerebinnedbyS/Noverthetime-integratedin-
2showsthenumberofGRBintervalsbothbeforeandafterthe tervaloftheGRBtoproducethetime-resolvedintervals.
cutswereappliedforthetime-integratedandtime-resolvedanal- 6. The individually significant S/N bins were then fit with a
ysis. BandfunctionfromaLow-EnergyThreshold(LET)to∼40
MeV.
7. TheLETswereselectedtobe15,20,25,30,50,100keV.
3.6. Varianceofα
8. Thefitfunctionwasthenextrapolatedto8keV.
Another indication that a time interval in a burst had a low- 9. Theresidualsweresummedbetween8keVandtheLETfor
energy spectral deviation was a variation in α when only the eachdetectorandaveragedacrossdetectors.
LET is changed. For each time section, each α parameter ob- 10. Simulations were then run and fit with a Gaussian distribu-
tainedfromusingdifferentLETswascompared.Ifthevalueof tion.
α remains consistent across all different LETs in a spectrum, it 11. Thesignificanceofdeviationsinthedatawerethenquanti-
isassumedthatnosignificantlow-energydeviationsarepresent. fiedbythenumberofstandarddeviationsbywhichthedata
If variation in α of > 2 σ for different LETs was evident, the variedfromthemeanofthesimulateddistribution.
6
D.Tierneyetal.:Anomaliesinlow-energyGRBspectrawithFermiGBM
12. TheαparametersbetweendifferentLETsforatimeregion s), GRB081215A (T0+0 - T0+8.4 s). A large 28.2 σ excess
werecompared. (normalised) is noted in GRB090902B and is a clear outlier
13. If a time region had a significant residual deviation and a in Figure 4. GRB090424 has a deficit of 6.1 σ in the time-
variation in α while only changing the LET, it was consid- integrated spectrum which appears to be due to an anomalous
eredtobeaspectraldeviation. spectral feature near E . These two GRBs also show devia-
peak
tionsinthetime-resolvedspectralanalysis.GRB081215Aman-
ifests as a 6.7 σ deficit but does not show up significantly in
4. Results thetime-resolvedanalysis.Thisdeficitappearstobecausedby
strong spectral evolution throughout the GRB with E de-
Inthecontextofthiswork,spectraldeviationscanbeclassified peak
creasing from > 1 MeV in the first 1 s time bin to ∼ 50 keV
into two broad categories - excesses and deficits. The former
at the end of the 8 s emission region (a time-resolved analysis
occurwhenthereisanexcessofphotonswithrespecttotheex- byZhangetal.(2011)usingdifferentbinningshowssimilarre-
trapolatedfitfunctionbelowtheLET.These,forexample,could
sults).
beexplainedbyanadditionalcomponentsuchasapower-lawor
blackbodydominatingatlowenergies.Deficitsoccur,notdueto GRB090902Bmaybefitwithanintenseadditionalpower-
alackofphotonsatlowenergiesbutbecauseαisflatterthanthe law component throughout the entire burst causing it to devi-
actualdataatlowenergies.Thiscanbecausedbythepresence atestronglyinthetime-integratedspectrum(Abdoetal.2009a;
of an additional component, such as a blackbody, between the Zhangetal.2011).GRB090424hasanintenseadditionalcom-
LETandE resultinginashallowerαandadeficitofphotons ponent, which can be modelled as a blackbody throughout the
peak
relativetotheextrapolatedfunctionatlowenergies. burstwhichalsomanifestsasadeviationinthetime-integrated
results.Additionalemissionprocessesandcomponentsmayal-
waysbepresentinallGRBsbuttheinstrumentmaynotbesen-
4.1. HypothesisTesting sitiveenoughtoconfidentlydetectthem.Acloserinspectionof
each GRB using time-resolved analysis is required to differen-
Two methods were used to test for deviations at low energies.
tiate between artifacts in the spectra and real additional time-
Thefirstinvolvescomparingtheresidualsfromthedatatothose
dependentspectralfeatures.
fromsimulationswhereaBandfunctionistheassumedbestfit.
TheresultsinFigure5showtwoGRBs,onewithandonewith-
outdeviationsusingthetime-integratedresults.Thenullhypoth-
4.3. Time-ResolvedSpectra
esis for these simulations is that if the data are consistent with
the simulated distribution, the extrapolated spectrum is consis-
Atime-resolvedanalysiswasperformedon672spectrawithan
tentwithaBandfunction. S/Nof25σand309spectrawithanS/Nof50σwhichresulted
insignificantfeaturesinfivebursts.GRB090323,GRB090424,
The second method involves looking at the variation in α
GRB090820, GRB090902B and GRB090926A contained at
whileonlychangingtheLETforthetime-resolvedanalysis.The
leastonetemporalintervalwheresignificantdeviationswerede-
changesinαforoneGRBwhichisconsistentwithaBandfunc-
tected.Simulationswereperformedontherelevanttimeregions
tion and one which is not consistent are presented in Figure 6.
to ensure that they were significant. The properties of the five
TwoGRBs,GRB090618andGRB091024showedvariationsin
burstsarepresentedinTable3.
αbutdidnothavelargedeviationsinthelow-energyresiduals.
Out of the five bursts observed with low-energy deviations,
Theparameterαvariedoutside2σduringonetemporalinterval
fourareinthetopfivemostfluentGRBsfromthefirst2years.In
foreachoftheGRBs.Theseintervalswereinvestigatedfurther
thecaseofthehighestfluenceevent(GRB090618withafluence
andwerediscountedduetoalong,weakintervalinGRB091024
of2.68×10−4±4.29×10−7erg/cm2(Paciesasetal.2012)),only
andonlyhavingasingle‘good’detectoravailableforanalysisin
oneNaIwasavailableforanalysisduetothesource-instrument
GRB090618.
geometry. It cannot be claimed with certainty that the result is
stableonthebasisofone‘good’NaIdetector.Ananalysisofthis
Simulations were performed for GRBs that initially pre-
burst with the combined GBM-Swift data did not significantly
sentedstrongdeviationsinthetime-integratedresults.Thesede-
demonstratetheneedforanadditionalcomponentintheprompt
viationswerethenquantifiedbycomparingthedatatothesim-
gamma-rayemission(Pageetal.2011).
ulateddistributionforthatGRBandwerediscardedasachance
occurrence if consistent with the simulations. Otherwise it was Asspectralevolutioncanleadtoanomalousbehaviour,even
classifiedasaspectraldeviation.AsampleofGRBsweresim- inthetime-resolvedanalysis,lightcurveswereexaminedforany
ulatedthatdidnotshowstrongdeviationstoensuretheveracity notablefeaturesduringatimeintervalwhichcontainedadevia-
ofthenullhypothesis. tion.Tolimittheeffectoftemporalbinning,severalfurtherfits
Simulations could not be performed for time-resolved re- wereperformedbychangingthelengthoftheintervalbyupto±
gions of all GRBs so the combination of the two methods out- 50%andbyshiftingtheintervalupto50%.Iftheeffectswere
linedpreviously(summingthelow-energyresidualsandcheck- stillsignificantafterthisfurtheranalysisitisclaimedasadetec-
ing the change in α between different LETs) present the final tionofanadditionalspectralfeature.Theseintervalswerethen
criterion by which a deviation is tested. If a deviation passed fit with more complex models and the results are summarised
bothteststhenitwasclassifiedasarealspectralanomaly. in Table 4. There is a notable improvement in the fit statistic
whenaBandfunctionisfitwithanadditionalcomponent.This
providesevidencethatadditionalspectralfeaturesareneededto
4.2. Time-IntegratedSpectra accuratelymodelthesetimeintervals.Thesesolutionshowever
arenotuniqueandotherspectralmodelsmayhaveasimilaror
Out of the sample of 45 GRBs tested, a significant devi-
betterfitstatistic.
ation in the time-integrated results was found in 3 cases
GRB090902B(T0+0-T0+25s),GRB090424(T0+0-T0+6.4
7
D.Tierneyetal.:Anomaliesinlow-energyGRBspectrawithFermiGBM
140 160
LET = 25 keV LET = 25 keV
120 140
120
100
100
80
80
60
60
40
40
20 20
0 0
15 10 5 0 5 10 15 30 20 10 0 10 20
Summed Sigma Residuals Summed Sigma Residuals
(a) (b)
Fig.5.Histogramsofsummedresidualsbelow25keVobtainedfromtime-integratedsimulationsofaperfectBandfunction.The
dark line shows the value of the data for that GRB. a) The value of the deviation for GRB080817A is -3.6. The mean and the
standarddeviationofthesimulateddistributionare-1.1and3.4respectively.Inthiscasethedatadoesnotdeviatesignificantlyfrom
thesimulateddistributionandhasanormaliseddeviationof-0.7σ.b)ThevalueofthedeviationforGRB090424is-25.8compared
tothesimulateddistributionwithameanandstandarddeviationof0.4and4.3respectively.Thedatadeviatessignificantlyfromthe
simulateddistributionwithanormaliseddeviationof-6.1σ.
Table3.PropertiesoftheGRBsfoundtohavestrongspectraldeviationsinthetime-resolvedanalysisatlowenergies.
GRB Fluencea FluenceRanka T90a Deviationb Localisationc
10-1000keV first2years 50-300keV
(erg/cm2) (s)
GRB090323 1.2×10−4±1.7×10−7 5 135.2±1.5 +4.9σ(Excess) SwiftXRT1
GRB090424 4.6×10−5±3.9×10−8 19 14.1±0.3 -6.2σ(Deficit) SwiftXRT2
GRB090820 1.5×10−4±1.8×10−7 3 12.4±0.2 -5.5σ(Deficit) GBM3
GRB090902B 2.2×10−4±3.2×10−7 2 19.3±0.3 +25.0σ(Excess) SwiftXRT4
GRB090926A 1.5×10−4±3.4×10−7 4 13.8±0.3 +6.5σ(Excess) SwiftXRT5
Notes. (a)FromPaciesasetal.(2012).(b)Thedeviationsfromtheextrapolatedfittingmethodhavebeennormalisedbythestandarddeviation
fromthesimulateddistributionsforeachindividualburstsection.(c)TheSwiftXRTlocationissufficientlyaccuratetomodeltheresponseofthe
instrumentforspectralanalysis,evenincaseswhereamoreaccuratelocationisknown.
References.(1)Kenneaetal.(2009);(2)Cannizzoetal.(2009);(3)Connaughton(2009b);(4)Kennea&Stratta(2009);(5)Vetereetal.(2009)
4.4. IndividualBurstswithanomalouslow-energyspectra 2009).Theenergeticsandhostgalaxypropertiesofthisburstare
describedinMcBreenetal.(2010).
SignificantanomalousspectralbehaviourwithrespecttoaBand Afterperformingaspectralanalysisonthisburstanexcessis
function was found in time intervals of five bursts. The spectra detectedintwooutofnineintervals(52-60s,60-66srelative
werefurtherinvestigatedbyfittinganumberofspectralmodels the trigger time). The spectral analysis for the interval with the
as presented in Table4. The individual events are described in strongestlow-energydeviationispresentedinFigure7andTable
moredetailbelow. 4. An improvement in the spectral residuals is notable when a
Band+blackbodyorBand+power-lawfitisperformedovera
singleBandfit.
4.4.1. GRB090323
GRB090323 was detected by Fermi GBM (van der Horst &
Xin 2009) and LAT which caused the satellite to perform an
4.4.2. GRB090424
autonomous repoint to the source location (Ohno et al. 2009).
ThelightcurvefromGBMNaIdetectorn9isdisplayedinFig.7 ThepromptemissionfromGRB090424wasdetectedbySwift-
and consists of multiple distinct pulses over ∼ 150 s. Multi- BAT, XRT and UVOT which provided an arcsecond localisa-
wavelength observations were aided by an on-ground location tiontothefollow-upcommunitywithinminutes(Cannizzoetal.
provided by the LAT and X-ray observations from Swift-XRT 2009).ThepromptemissionwasalsoobservedbyFermiGBM
(Kennea et al. 2009). Follow-up observations report a spectro- (Connaughton2009a)andSuzakuWAM(Hanabataetal.2009).
scopic redshift of z = 3.57 (Chornock et al. 2009a) and the af- Multiple optical telescopes observed the optical afterglow (e.g.
terglow was also detected at radio wavelengths (Harrison et al. (Olivaresetal.2009b))andaspectroscopicredshiftofz=0.544
8
D.Tierneyetal.:Anomaliesinlow-energyGRBspectrawithFermiGBM
0.2
0.3
α 0.4
0.5
0.6
10 20 30 40 50 60
Low-Energy Threshold (keV)
(a) (b)
0.4
0.6
α
0.8
1.0
1.2
10 20 30 40 50 60
Low-Energy Threshold (keV)
(c) (d)
Fig.6. Change in α between different LETs. The S/N temporal selections in GRB090102 and GRB090902B are presented in a)
andc).Thespectralparameterαispresentedwith2σerrorbarsinb)andd)forLETvaluesfrom15keVto50keV.αisconsistent
at the 2 σ level as the LET is varied for the time interval selected in GRB090102 and is not consistent for the time interval in
GRB090902B.
was obtained by Gemini-South (Chornock et al. 2009b). Radio ismostprominentbetween2-3srelativetothetriggertime.Six
observationsdetectedabrightradioafterglow(Chandra&Frail 1sintervalsbetween0-6swereanalysedandastrongdeviation
2009) and a disturbance in Earth’s ionosphere was detected by ispresentintheresidualsinthethe2-3sregion.Littlespectral
monitoringverylowfrequencyradiowaves(Mondaletal.2011; evolution is present throughout the burst, and as this feature is
Chakrabartietal.2010). observablethroughouttheburstitislesslikelytobecausedby
The main emission period of this GRB is relatively short spectralevolution.
compared to other bursts in the high fluence sample and has
an extremely high peak flux of 110 photons / s (Paciesas et al.
2012). The burst has at least 3 pulses in rapid succession last-
4.4.3. GRB090820
ing ∼ 6 s and lower level emission that continues up to 20 s
after the trigger time. A significant low-energy deficit was ob- GBMdetectedGRB090820andissuedanautomatedrepointre-
servedinthetime-integratedandtime-resolvedspectralanalysis questtoFermi,howevertheLATcouldnotobservetheburstdue
of GRB090424. This apparent deficit is caused by the unusual toEarthavoidanceconstraints(Connaughton2009b).TheRT-2
spectral shape, which is shown in Figure 8. The spectrum has ExperimentonboardtheCORONAS-PHOTONsatellitealsoob-
2 distinct spectral breaks, one in the usual energy range where servedthepromptemission(Chakrabartietal.2009).Nofurther
a break due to E is to be expected but also another lower follow-up of the burst occurred and thus the best location for
peak
energy break in the spectrum. A Band + blackbody fit with a theburstisobtainedfromGBM.Thebursttriggeredonaweak
blackbodykTof∼9keVcanbefitthroughouttheburstwhich precursorwhichwasexcludedfromthespectralanalysis.
9
D.Tierneyetal.:Anomaliesinlow-energyGRBspectrawithFermiGBM
Table4.Time-resolvedspectralparametersfrommodelfitstoGRBtimeintervalswheresignificantdeviationsweredetected.
GRB Interval∗ Detectors Model Epeak α β kT Index C-Stat/DOF
GRB090323 60.4:66.6 n9+nb+b1 Comp 535.70+33.00 -0.83+0.03 - - - 410.14/353
Band 532.00+−3259..3900 -0.83+−00..0033 -2.92+0.37 - - 409.25/352
Band+BB 426.00+−2361..2900 -0.52+−00..0083 -3.−040.120 4.99+0.53 - 373.13/351
Band+PL 409.20+−2283..8500 -0.28+−00..1047 -3.002 −0.52- -1.75+0.08 377.69/351
−26.30 −0.13 −0.14
GRB090424 2.3:3.3 n6+n7+n8 Comp 169.50+4.61 -0.87+0.03 - - - 738.26/596
+nb+b1 Band 153.00−+47..4001 -0.80−+00..0034 -2.81+0.17 - - 729.21/595
Band+BB 176.90+−107..9303 -0.48+−00..1024 -3.10+−00..2233 9.20+0.55 - 662.65/593
Band+PL 153.40−+8.96.144 -0.80−+00..1015 -2.82+−00..2523 −0.44- 0.16+INF 729.25/593
−9.44 −0.00 −0.22 −0.00
GRB0908201 31.7:36.9 n2+n5+b0 Comp 267.10+2.36 -0.58+0.01 - - - 896.24/357
Band 221.60+−32..4365 -0.45+−00..0011 -2.63+0.04 - - 667.45/356
Band+BB 317.20+−93..2445 -0.71−+00..0012 -3.37+−00..1084 28.90+0.91 - 600.59/354
Band+PL −9.44- −0.02- −0.26- −0.95- - -
GRB090902B3 9.7:10.6 n0+n1+b0 Comp 1695.00+74.50 -1.14+0.01 - - - 1178.7/356
Band 1657.00+−9875..3300 -1.14+−00..0011 -3.002 - - 1198.0/356
Band+BB 917.30+−3676..7800 -0.51+−00..0040 -3.002 5.01+0.15 - 489.48/354
Band+PL 782.00+−3334..5400 0.20+−00..1014 -3.002 −0.16- -2.02+0.04 419.13/354
−27.70 −0.12 −0.06
GRB090926A 9.5:10.5 n3+n6+n7 Comp 331.90+10.10 -0.95+0.02 - - - 639.49/478
+b1 Band 311.40−+911.4.690 -0.93−+00..0022 -2.64+0.13 - - 628.82/477
Band+BB 392.00+−4141..5500 -1.17−+00..0025 -2.35+−00..1119 48.99+3.54 - 592.81/475
Band+PL 265.70+−1411..7300 -0.46+−00..0094 -3.35+−00..4114 −2.81- -1.73+0.04 583.77/475
−11.20 −0.09 −1.09 −0.04
Notes. *) Measured in seconds since trigger time (T0). 1) The Band+PL fit for GRB090820 did not converge to physical parameters and is
omitted.2)βwasfrozento-3.00iftheuncertaintieswereunconstrained.3)AlthoughmanyintervalshavesignificantexcessesinGRB090902B,
onlyoneispresentedaboveforillustrativepurposes(SeealsoAbdoetal.2009a;Pe’eretal.2012;Rydeetal.2011).
This burst presents a low-energy deficit relative to a Band inTable4.Anadditionalpower-lawcomponentwasthebestfit
functionwhichiscausedbytheBandfunctionattemptingtofit modeltothedataoutofthemodelstested.Theadditionalpower-
datathathasabroaderpeakrelativetotheexpectedfunctionfit. lawindexfromanalysingtheGBMdataonlyinthisregion,α=
Thirteen time intervals between 29 - 45 s from the trigger time -2.00+0.04,isconsistentwiththepower-lawindexobservedbya
wereanalysedwithfive1sintervalsbetween31-36sexhibit- joint−fi0t.0o6ver the entire spectral range of GBM and LAT (α = -
ingmarginalevidenceofdeviations.AsthiswastheonlyGRB 1.98+0.02).ComplexspectralmodelstothisGRBarefitinFigure
−0.02
withconsecutivemarginalevidencefordeviations,thesetimein- 10. Numerous models have been presented in the literature to
tervalswereanalysedtogetheranddisplayastrongdeviationin explain the spectral features in this burst (Liu & Wang 2011;
the spectral data. By adding a blackbody component the Band BarniolDuran&Kumar2011;Pe’eretal.2012).
function can account for the broader spectral peak as shown in
Figure9.Asanadditionalpower-lawcouldnotaccountforde-
viationsatlowerenergies,thisprovidesevidencethatsomead-
ditional spectral property is occurring in the mid-energy range.
4.4.5. GRB090926A
Spectralanalysiscomparingstandardmodelstomoreadvanced
fittingtechniques,suchassynchrotronmodelshavebeencarried
outbyBurgessetal.(2011)forthisburst. Detected by GBM (Bissaldi 2009) and LAT (Uehara et al.
2009),follow-upobservationsofthisGRBweremadebySwift-
XRT(Vetereetal.2009)andSwift-UVOT(Gronwall&Vetere
2009).ThepromptemissionwasalsoobservedbySuzakuWAM
4.4.4. GRB090902B (Noda et al. 2009) and Konus-Wind (Golenetskii et al. 2009).
A redshift of z = 2.1062 was obtained by the Very Large
GRB090902B (Bissaldi & Connaughton 2009) was an ex- Telescope(VLT)X-Shooterspectrograph(Malesanietal.2009).
tremely bright GRB with a bright additional component ob- Skynet/PROMPT also observed the optical afterglow (Haislip
served across 8 orders of magnitude in energy (Abdo et al. etal.2009).
2009a). The GRB was initially detected by both instruments
GRB090926A has also been observed in joint GBM and
on Fermi, GBM and the LAT (Bissaldi & Connaughton 2009;
LATspectralfittingtonecessitateanadditionalpower-lawcom-
de Palma et al. 2009). The prompt emission was also detected
ponent over a wide spectral range (Ackermann et al. 2011).
bySuzakuWAM(Teradaetal.2009).ATargetofOpportunity
Thirteentimebinswereanalysedbetween0-17swithonetime
(ToO)wasissuedtoSwiftwhichobservedanuncatalogedsource
binshowingasignificantexcess.Theexcessisobservedduring
withintheLATerrorradius(Kennea&Stratta2009).Numerous
ashortsharpspikeinthelightcurvewhichisshowninFigure11.
optical follow-up observations were made (e.g. Olivares et al.
2009a) resulting in a redshift of z = 1.822 (Cucchiara et al. Whenanadditionalpower-lawisfitwithaBandfunction,thefit
statistic shows a large improvement. The additional power-law
2009).Thesourcewasalsoobservedintheradio(vanderHorst
indexα=-1.73+0.04,obtainedbyusingGBMdataonly,isagain
etal.2009). −0.04
consistentwiththepower-lawobservedintheLATathighener-
AsignificantexcesswasobservedinthisGRBinnumerous giesα=-1.71+0.02(Ackermannetal.2011).
intervals (ten 1-second intervals between 6 - 16 s relative the −0.05
triggertime).Theregionwiththestongestdeviationispresented
10