Long-term lightcurves from combined unified very high energy γ -ray data
M. Tluczykont, E. Bernardini, K. Satalecka, R. Clavero, M. Shayduk, O. Kalekin
aa r X i v : . [ a s t r o - ph . H E ] O c t Astronomy&Astrophysicsmanuscript no. tluczykont˙lc c (cid:13)
ESO 2018October 16, 2018
Long-term lightcurves from combined unified very high energy γ -ray data M. Tluczykont , , E. Bernardini , K. Satalecka , R. Clavero , , M. Shayduk , , and O. Kalekin , DESY Platanenallee 6, 15738 Zeuthen, Germanye-mail: [email protected] Now at Universit¨at Hamburg, Luruper Chaussee 149, 22761 Hamburg, Germany Now at Isaac Newton Group of Telescopes, Apartado de correos 321, E-38700 Santa Cruz de la Palma, Canary Islands, Spain Now at Max Plank Institut f¨ur Physik, F¨ohringer Ring 6, 80805 Muenchen, Germany Now at Erlangen Centre for Astroparticle Physics, Erwin-Rommel-Str. 1, 91058 Erlangen, GermanyReceived ; accepted
ABSTRACT
Context.
Very high-energy (VHE, E >
100 GeV) γ -ray data are a valuable input for multi-wavelength and multi-messenger (e.g. com-bination with neutrino data) studies. Aims.
We aim at the conservation and homogenization of historical, current, and future VHE γ -ray-data on active galactic nuclei(AGN). Methods.
We have collected lightcurve data taken by major VHE experiments since 1991 and combined them into long-termlightcurves for several AGN, and now provide our collected datasets for further use. Due to the lack of common data formats inVHE γ -ray astronomy, we have defined relevant datafields to be stored in standard data formats. The time variability of the combinedVHE lightcurve data was investigated, and correlation with archival X-ray data collected by RXTE / ASM tested.
Results.
The combination of data on the prominent blazar Mrk 421 from di ff erent experiments yields a lightcurve spanning morethan a decade. From this combined dataset we derive an integral baseline flux from Mrk 421 that must be lower than 33 % of the CrabNebula flux above 1 TeV. The analysis of the time variability yields log-normal flux variations in the VHE-data on Mrk 421. Conclusions.
Existing VHE data contain valuable information concerning the variability of AGN and can be an important in-gredient for multi-wavelength or multi-messenger studies. In the future, upcoming and planned experiments will provide moredata from many transient objects, and the interaction of VHE astronomy with classical astronomy will intensify. In this con-text a unified and exchangeable data format will become increasingly important. Our data collection is available at the url: http://nuastro-zeuthen.desy.de/magic experiment/projects/light curve archive/index eng.html . Key words.
Gamma-rays: observations – Galaxies: active – Galaxies: individual: Mrk 421 – Galaxies: individual: Mrk 501
1. Introduction
The broad-band emission observed from active galactic nuclei(AGN) spans the complete electromagnetic spectrum from radioto VHE (very high-energy, E >
100 GeV) γ -rays. Since the dis-covery of the blazar Mrk 421 in the VHE regime (Punch et al.1992), many new detections of AGNs by di ff erent experimentshave been reported (see, e.g. Horns 2008; Hinton & Hofmann2009, for reviews). Strong flux variations on di ff erent timescaleswere observed from many AGNs. Variability in the VHE γ -rayregime was measured down to the minute timescale (MAGICCollaboration et al. 2008; Aharonian et al. 2007). Many AGNlightcurves were produced by di ff erent γ -ray experiments fromobservations of flaring states, dedicated monitoring programs, orfrom joint campaigns with other experiments. Long-term γ -raylightcurves are valuable to address open questions concerningAGNs, such as possible periodicities (see Thieler et al. 2010, ananalysis using the data collection of the present publication), thelog-normality of their flux state distribution, the nature of theradiation mechanism of AGNs, or the estimation of their duty-cycle.A log-normal behavior can be indicative of a multiplica-tive process of the underlying mechanism governing the vari-ability of the object. In the case of AGNs, this could be evi- dence for a connection of the observed emission to accretiondisk activity (see Giebels & Degrange 2009, and referencestherein). Previously, log-normal flux variations were reportedin the X-ray band from two objects: the narrow-line Seyfert 1galaxy IRAS 13244-3809 (Gaskell 2004) and BL Lac (Giebels& Degrange 2009). In VHE data, log-normal variability was ob-served from a high flux state of the BL Lac object PKS 2155-304(Degrange et al. 2008).The nature of the AGN radiation mechanism in the VHEregime, i.e. whether the observed radiation has a leptonic orhadronic origin, still remains ambiguous. The detection of neu-trinos from these objects would prove the existence of a hadroniccomponent. In the framework of multi-messenger strategies(e.g., combining electromagnetic with neutrino data), the phe-nomenology of lightcurves in the electromagnetic wavelengthband can give valuable input. The Neutrino triggered Targetof Opportunity (NToO) program (Bernardini 2005; Ackermannet al. 2008) is based on the idea of neutrino events from a vari-able object (single events or multiplets) that are used to triggerVHE monitoring of the same object. Coincidences between neu-trino triggers and γ -ray flux high states, which occur more oftenthan expected from random coincidence with atmospheric neu-trinos, would be evidence of a hadronic component of the VHE γ -ray signal and a cosmic origin of the neutrinos. Long-term M. Tluczykont et al.: Long-term γ -ray lightcurves γ -ray lightcurves can be used to estimate the AGN dutycycle,which is an important input parameter for such analyses.
2. Lightcurve data
Considering the heterogeneous nature of historical and presentVHE data comprising di ff erent file and content formats fromdi ff erent experiments, a common data format is desirable.Moreover, as described above, VHE astronomy has alreadystarted to interface strongly with di ff erent fields of classical as-tronomy. In view of an e ff ective exchange and di ff usion of VHEdata within the astronomy and astroparticle community, our stor-age strategy is to use an ascii file format (SLF) and the widelyused FITS and VOTable file formats (see following section).The use of a standard data format will also become in-creasingly important in the framework of the next-generationCherenkov telescope systems, such as the upcoming CTA(Cherenkov Telescope Array) experiment (see e.g. Martinez2008). In the CTA era, VHE astronomy will intensify multi-wavelength (and multimessenger) interactions with other fields.Furthermore, a standard data format is essential for running anexperiment like an observatory, and make it open to the wholeastronomy community.Publicly available lightcurve data from 1992 until today werecollected from the Whipple (Kerrick et al. 1995; Schubnell et al.1996; Buckley et al. 1996; Maraschi et al. 1999a), HEGRA(Aharonian et al. 1999b,a; Krawczynski et al. 2001b; Aharonianet al. 2001, 2002, 2003, 2004; Kestel 2002), CAT (Piron 2000;Piron et al. 2001), H.E.S.S. (Aharonian et al. 2005, 2006),MAGIC (Albert et al. 2008; Donnarumma et al. 2009), andVERITAS (Rebillot et al. 2006; Donnarumma et al. 2009) exper-iments. We are also working on collecting more data from theseand other experiments such as the Crimean GT-48 Cherenkovtelescope (e.g. Neshpor et al. 2007), the Tibet air-shower ar-ray (e.g. Amenomori et al. 2003), or the Patchmarhi CherenkovTelescope Array (e.g. Gupta et al. 2008). Di ff erent formats and standards are used by di ff erent experi-ments. We use simple directly usable ascii tables and the stan-dard astronomy file formats FITS and VOTable to store the col-lected, combined data. In the present work, the types and unitsof the FITS / VOTable datafields (i.e. columns) follow a simplelightcurve format (hereafter SLF) given in Table 1. The data arealso provided in ascii file form following the SLF definition.These files are referred to as slf-files. The tables contain rowsof (typically nightwise) integral flux-measurements and furtherinformation necessary for subsequent analyses. The units of thedatafields are defined within the FITS / VOTable files and followthe conventions defined in Table 1 within the slf-files (ascii).The SLF datafield format definition provides the means to e ff ec-tively combine heterogeneous datasets. In case further informa-tion than included in columns 1 to 12 are necessary, additionalentries are foreseen. For example, to define a spectral shape dif-ferent from the standard pure powerlaw, the spectral parameter-ization together with the additional parameters must be definedas comments within the data files. In case of lightcurve data onlycontaining information on a few datafields, the flexibility of thebinary FITS and VOTable file format allows minimizing the datavolume. For example, if only the observation date and the flux value are known for the total lightcurve data of an object, thebinary file can only contain MJD start and F , along with the def-inition of units used. While this example represents the simplestcase of a homogeneous dataset, in practice, the heterogeneouslightcurve data combinations from di ff erent experiments can in-clude information on a specific datafield for one subset of thecombined data (i.e. for one experiment), and not for another sub-set. In this (most common) case, when information is not avail-able for datafields of some of the measurements, the followingconventions for default datafield values apply. – If no start and end times are known, then
MJD start is set tothe same value as
MJD mid , i.e. the middle of the exposuretime, and
MJD end to -1. In this case the duration might stillbe given. – In some cases only the MJD of the observations day isknown. Then,
MJD mid is the MJD of the observation day. – Most energy spectra can be described by a pure powerlaw(di ff erential spectral index α ) or a powerlaw with an expo-nential cuto ff (cuto ff energy E cuto f f ). Additional fields areforeseen for cases with di ff erent parameterizations of the en-ergy spectrum. If such a di ff erent parameterization is given,a defining formula containing the relevant fields has to bewritten in the comment section of the lightcurve file. – The flux-flag ( ffl ag) is a one-character flag indicatingwhether the integral flux F is an upper limit ( ffl ag = ’ < ’)or a flux measurement ( ffl ag = ’ = ’). – In the present work, we chose Crab Nebula flux units for theobserved integral flux F above the energy threshold E thr . – For any other entry, the value -1 means that no informationis available.
For a combination of lightcurves from di ff erent experiments, themeasured integral flux values must be converted to a commonenergy threshold. For the results in this paper, we chose an en-ergy threshold of 1 TeV. The conversion of integral fluxes in Crabunits to the same energy threshold requires knowledge of the en-ergy spectrum of the considered object and of the Crab Nebula asmeasured by the same experiment. The di ff erential energy spec-trum of the Crab Nebula in the energy range covered by mostVHE γ -ray experiments is described well by a pure powerlawof the form φ Crab ( E ) = φ (0) Crab · ( E / T eV ) − Γ Crab in the energy rangefrom 100 GeV to few tens of TeV. Deviations from the pure pow-erlaw form at both ends of this energy range (as expected in theframework of the inverse Compton scenario) were not taken intoaccount here. Extrapolation to energies below 100 GeV requiresa full parameterization of the spectral energy distribution of theCrab Nebula (e.g. Aharonian et al. 2004). An integral flux inCrab units above the energy threshold E thr is given by F ( E > E thr ) = R ∞ E thr dE φ ( E ) R ∞ E thr dE φ Crab ( E ) = R ∞ E thr dE φ ( E ) F Crab ( E > E thr ) (1)where φ ( E ) and φ Crab ( E ) are the di ff erential energy spec-tra of the source and of the Crab Nebula (in units of pho-tons TeV − cm − s − ). The corresponding integral fluxes above agiven energy threshold are denoted with F ( E > E thr ). In thecase of a pure powerlaw di ff erential energy spectrum φ ( E ), theconversion of an observed integral flux F ( E > E ) to a giventhreshold energy E thr is given by F ( E > E thr ) = E thr E ! − Γ+ F ( E > E ) F Crab ( E > E thr ) , (2) . Tluczykont et al.: Long-term γ -ray lightcurves 3 Datafield Name TTYPE TUNIT Explanation MJD mid mjd mid exp MJD center of observation exposure2
MJD start mjd start MJD start of observation3
MJD end mjd end MJD end of observation4 F int flux Crab units Observed integral flux5 ∆ F stat sigma int flux stat Crab units Statistical error on F ∆ F sys sigma int flux sys Crab units Systematical error on F α alpha none Di ff erential spectral index8 ∆ α stat sigma alpha stat none Statistical error on α ∆ α sys sigma alpha sys none Systematical error on α E thr e thr TeV Energy threshold in TeV11 E cutof f e cut TeV Cuto ff energy in TeV12 Experiment experiment none Experiment string13 Duration duration hours Duration of observation14 Reference reference none Reference string (pref. ADS format)15 Flux-flag ffl ag none flux-measurement: ’ = ’, upper limit: ’ < ’16 additional entry 1 add entry1 e.g., special parametrizations17 additional entry 2 add entry2... ... ... Table 1.
Datafield definition ∗ used for storage of lightcurve data in the FITS / VOTable file format. ∗ The used datafield definition is referred to as simple lightcurve format (SLF) in the text. TTYPE and TUNIT are the name and units used withinthe FITS file. where E is the energy threshold of the observation.The combined day-wise, integral flux lightcurve of theBL Lac object Mrk 421 above 1 TeV is shown in Figure 1. Thislightcurve includes all data we have collected so far, cover-ing an uprecedented 17-year time-span from 1992 to 2008. InFigure 2, the day-wise integral flux lightcurve of the BL Lac ob-ject Mrk 501 is shown. The inhomogeneity of data combined from di ff erent experimentsand the inclusion of data from the pioneering time of VHE γ -rayastronomy induces systematic errors that are not easily evalu-ated. Combination of data from many di ff erent experiments, dif-ferent time periods, partly inaccurate flux normalizations (i.e. γ -ray rate measurements) and energy thresholds can lead to sys-tematic errors in the relative flux normalizations. When com-bining data, the flux normalization depends on assumptions onthe spectral shape and the energy threshold. Here, limiting fac-tors can be experimental uncertainties (especially in old data)induced by poor knowledge of atmospheric conditions, aging de-tector e ff ects, and intrinsic spectral variability of the consideredobjects. We convert flux and rate measurements to Crab Nebulaflux units using Crab Nebula data taken by the same experimentas close in time as possible, therewith reducing systematic un-certainties induced by seasonal or instrumental variations.The error on the reconstructed spectral shape results in anerror on the flux state when converting to di ff erent thresholds.For example, when converting integral fluxes from 1 TeV to100 GeV, typical systematic errors on the spectral index of 0.1(pure powerlaw case) lead to a relative systematic error on theintegral flux of 30 % due to this extrapolation. The datafieldscontained in our FITS files provide all information needed to es-timate these systematic errors for conversion to di ff erent energythresholds. A conversion over a narrower energy range than usedin this example limits this systematic error to less than 30 %.The flux units used in some publications are given in un-calibrated units (e.g. counts per minute), i.e. as an instrument-specific γ -ray count-rate. The energy threshold is also unknownin some cases. For flux calibration of lightcurves from theWhipple experiment before 1998, we used Crab Nebula data taken by the same experiment as close in time as possible (seeHillas et al. 1998, and references therein). However, due to de-pendencies on zenith angle and weather conditions, normaliza-tion of rate measurements is very uncertain. We estimate thatthe systematic error induced when including rate measurementsis below 40 %. This form of relative systematic error can beavoided when restricting analysis to flux measurements, by re-stricting to data from a single experiment, or by intercalibratingthe di ff erent measurements when observations from di ff erent ex-periments overlap.
3. Discussion
Significant evidence of a correlated gamma- and X-ray emissionof blazars was presented in earlier studies (see e.g. Takahashiet al. 1996; Buckley et al. 1996; Maraschi et al. 1999b;Krawczynski et al. 2001a; Neshpor et al. 2007). Usually the re-ported correlation is linear, but in a few cases a quadratic re-lation between the fluxes in both bands was found (see e.g.Krawczynski et al. 2000). Such a correlation provides us withessential information on the underlying acceleration and emis-sion processes and is especially valuable for variability stud-ies. Very often the gamma- X-ray correlation is interpreted as astrong argument in favor of the so-called synchrotron-Comptonjet emission models in which the same population of ultrarel-ativistic electrons is responsible for production of both X-raysvia synchrotron radiation and TeV γ -rays via inverse Comptonscattering (Katarzy´nski et al. 2005). However, it can also be ac-commodated in the hadronic framework in particular in mod-els which assume that the observed γ -ray emission is a resultof interactions of accelerated protons and ambient gas or low-frequency radiation (Aharonian 2000; M¨ucke et al. 2003).Using contemporanous X-ray data extracted from the RXTE / ASM database web interface at MIT , we calculated cor-relation coe ffi cients between the VHE and X-ray bands. For eachVHE measurement, an average ASM count rate was calculatedcentered on the VHE observation date ( ± http: // xte.mit.edu / ASM lc.html
M. Tluczykont et al.: Long-term γ -ray lightcurves MJD [days] I n t e g r a l F l u x > T e V [ C r a b ] Mrk 421 longterm VHE lightcurveHEGRA CTSHEGRA CT1H.E.S.S.MAGICWhipple/VERITASCAT zoom: 2000/2001
Fig. 1.
Long-term lightcurve of Mrk 421 (day-wise integral flux). Data from the major γ -ray telescopes were combined and normalized to thesame energy threshold (1 TeV) and converted to Crab units (see text). A zoom into the period of strong activity (2000 / MJD [days] I n t e g r a l F l u x > T e V [ C r a b ] Mrk 501 longterm VHE lightcurveHEGRA CTSHEGRA CT1MAGICWhipple/VERITAS
Fig. 2.
Long-term lightcurve of Mrk 501. Available data were combined and normalized. Shown are the day-wise integral fluxes above 1 TeV inunits of the Crab Nebula flux. ment, these data yield correlation coe ffi cients (Spearman rank)of 0.65 for Mrk 421 and 0.68 for Mrk 501. In both cases, theprobability of an uncorrelated system producing datasets withsimilar Spearman rank correlation coe ffi cients is less than 10 − .The flux measurements in both wavelength bands are not strictlysimultaneous. However, they represent a measure for the aver-age daily flux state of the objects, and the observed behavior isconsistent with correlated average VHE / X-ray daily flux levelsmodulated by shorter term flux variations.In Figure 3, the distribution of the VHE flux states ofMrk 421 (also see Figure 1) is shown. Each entry of this his- togram represents a snapshot flux measurement of Mrk 421 asgiven by the combined dataset. All individual observed inte-gral flux values were converted to flux values in units of theCrab Nebula flux and normalized to a common energy thresholdof 1 TeV. The observed energy spectra of the Crab Nebula andMrk 421 as observed by the individual experiments were takeninto account. The overall distribution can be described by an ex-ponential law (left panel) above a flux of a few tenths of Crab. Anexponential shape might indicate that the measured flux statesmainly reflect a stochastic outburst state of the object. As shownin the right panel of Figure 3, a better fit to the data is obtained . Tluczykont et al.: Long-term γ -ray lightcurves 5 / ndf χ ± ± -1.129 >1 TeV [Crab] VHE F ob se r va t i on s -2 -1 / ndf χ ± ± -1.129 / ndf χ Gauss
N 7.99 ± Gauss µ ± Gauss σ ± Ln N 3.27 ± Ln σ ± Ln µ ± >1TeV [Crab] VHE F ob se r va t i on s -2 -1
10 110 / ndf χ Gauss
N 7.99 ± Gauss µ ± Gauss σ ± Ln N 3.27 ± Ln σ ± Ln µ ± Fig. 3.
Distribution of VHE flux states of Mrk 421. An exponential function fit above a flux of 0.25 Crab, (avoiding detector threshold e ff ects) candescribe the data (left). The data are very well described by a fit of a Gaussian + log-normal distribution (right). when using the sum of a Gaussian and a log-normal distribution(Aitchison & Brown 1957; Limpert et al. 2001), as given by f ( x ) = N Ln x σ √ π exp − ( log ( x ) − µ ) σ ! . (3)The data were further divided into time intervals of equal length,each comprising 40 flux observations f i with statistical errors σ i . The excess variance σ xs = q N P Ni = ( f i − ¯ f ) − σ i (Vaughanet al. 2003) was calculated for each time interval. The ex-cess variance σ xs is a measure of the Poisson noise correctedrms value of the fluxes within the corresponding time inter-val. Figure 4 shows σ xs as a function of the average flux ¯ f within the interval. A clear proportionality is seen with σ xs ∼ (0 . ± . f i . A log-normally shaped distribution with a pro-portionality of the excess variance with the flux are evidence oflog-normal flux variations (Aitchison & Brown 1957).Below a flux level of few tenths of Crab, one might expecta low baseline flux level that can be described by a Gaussian.At these low fluxes, however, detector sensitivity threshold ef-fects become increasingly important, i.e. lower fluxes cannotbe detected significantly within the short observation windows.Therefore, the mean of the Gaussian fit (0.33 Crab flux units)must be treated as an absolute upper limit on the integral base-line flux above 1 TeV from Mrk 421. More information such as / Crabf / C r a b xs σ Fig. 4.
The excess variance σ xs as a function of the average flux withinequal-length intervals. A line fit corresponding to σ xs ∼ (0 . ± . f i is also given. the variability type (e.g. red noise / white noise / blue noise) orthe duty-cycle of the object can be extracted from the flux statedistributions. Previously, first steps in this direction were takenby Tluczykont et al. (2007); however, further investigations arenecessary using simulations and will be the topic of a subsequentpublication.With increasing exposure achieved by the major current gen-eration experiments more data will soon be available. As dis-cussed previously, one important aspect for long-term variabil-ity studies will be to carry out unbiased (random) observationsto avoid systematic selection e ff ects.
4. Summary & outlook
VHE γ -ray flux measurements of AGN have been collected fromobservations of di ff erent experiments since 1992. For the firsttime these data were combined into single long-term lightcurves,and are provided for further analysis in the standard FITS fileformat. The BL Lac object Mrk 421 yields the most extendeddataset with a combined lightcurve spanning more than a decade.The collected data is publicly available for download The observed flux states averaged over 12 h of Mrk 421and Mrk 501 above 1 TeV are consistent within statistical andsystematic errors with a linear correlation with daily averaged
RXTE / ASM X-ray count rates, with correlation coe ffi cients of0.65 and 0.68 respectively.The long-term flux state distribution of Mrk 421 can be welldescribed by the sum of a Gaussian and a log-normal distribu-tion. This behavior can be interpreted by the observation of abaseline flux and a stochastic flux state population governed byan underlying multiplicative process. In the framework of thisinterpretation, the combined dataset on Mrk 421 allowed settingan absolute upper limit on the baseline flux of Mrk 421 of 0.33Crab flux units above 1 TeV.Unbiased data from monitoring campaigns by ongoing ex-periments (for a review see Weekes 2008) and data from futureexperiments such as CTA will further extend the long-term cov-erage of variable objects. A unified standard data format willbecome increasingly important in the future, when more high-sensitivity VHE data becomes available, and when interfacingwith other fields of classical astronomy and astroparticle physics(multi-wavelength and multi-messenger) intensifies. We intend http://nuastro-zeuthen.desy.de/magic experiment/projects/light curve archive/index eng.html M. Tluczykont et al.: Long-term γ -ray lightcurves to store these future data as described in this work in standarddata formats using the introduced SLF field structure definition. Acknowledgements.
We thank all colleagues who provided data in electronicform to us. MT thanks Dr. B. Giebels and Prof. Dr. D. Horns for valuablediscussions. We thank Dr. B. Pence for helpful comments on an ealier draft.This research has made use of NASA’s Astrophysics Data System BibliographicServices.
References
Ackermann, M., Bernardini, E., Galante, N., & et al. 2008, in InternationalCosmic Ray Conference, Vol. 3, International Cosmic Ray Conference,1257–1260Aharonian, F., Akhperjanian, A., Barrio, J., et al. 2001, ApJ, 546, 898Aharonian, F., Akhperjanian, A., Beilicke, M., et al. 2002, A&A, 393, 89Aharonian, F., Akhperjanian, A., Beilicke, M., et al. 2004, ApJ, 614, 897Aharonian, F., Akhperjanian, A., Beilicke, M., et al. 2003, A&A, 410, 813Aharonian, F., Akhperjanian, A. G., Aye, K., et al. 2005, A&A, 437, 95Aharonian, F., Akhperjanian, A. G., Barrio, J. A., et al. 1999a, A&A, 349, 29Aharonian, F., Akhperjanian, A. G., Bazer-Bachi, A. R., et al. 2007, ApJ, 664,L71Aharonian, F., Akhperjanian, A. G., Bazer-Bachi, A. R., et al. 2006, A&A, 457,899Aharonian, F. A. 2000, New A, 5, 377Aharonian, F. A., Akhperjanian, A. G., Barrio, J. A., et al. 1999b, A&A, 342, 69Aitchison, J. & Brown, J. A. C. 1957, The Lognormal Distribution, cambridgeUniversity Press: CambridgeAlbert, J., Aliu, E., Anderhub, H., et al. 2008, ApJ, 674, 1037Amenomori, M., Ayabe, S., Cui, S. W., et al. 2003, ApJ, 598, 242Bernardini, E. 2005, in Proceedings Towards A Network Of CherenkovDetectors VII, ed. B. Degrange, Palaiseau, 307–315Buckley, J. H., Akerlof, C. W., Biller, S., et al. 1996, ApJ, 472, L9 + Degrange, B., Superina, G., Giebels, B., & Volpe, F. 2008, in Blazar Variabilityacross the Electromagnetic SpectrumDonnarumma, I., Vittorini, V., Vercellone, S., et al. 2009, ApJ, 691, L13Gaskell, C. M. 2004, ApJ, 612, L21Giebels, B. & Degrange, B. 2009, A&A, 503, 797Gupta, A. C., Acharya, B. S., Bose, D., Chitnis, V. R., & Fan, J. 2008, ChineseJ. Astron. Astrophys., 8, 395Hillas, A. M., Akerlof, C. W., Biller, S. D., et al. 1998, ApJ, 503, 744Hinton, J. A. & Hofmann, W. 2009, ARA&A, 47, 523Horns, D. 2008, in Reviews in Modern Astronomy, ed. S. R¨oser, Vol. 20, 167– + ,arXiv:astro-ph / ff use X-ray Background, 599,694Krawczynski, H., Sambruna, R., Kohnle, A., et al. 2001b, ApJ, 559, 187Limpert, E., Stahel, W., & Abbt, M. 2001, BioScience, 51 (5), 341352MAGIC Collaboration, Albert, J., Aliu, E., et al. 2008, Physics Letters B, 668,253Maraschi, L., Fossati, G., Tavecchio, F., et al. 1999a, ApJ, 526, L81Maraschi, L., Fossati, G., Tavecchio, F., et al. 1999b, Astroparticle Physics, 11,189Martinez, M. 2008, in American Institute of Physics Conference Series, Vol.1085, American Institute of Physics Conference Series, ed. F. A. Aharonian,W. Hofmann, & F. Rieger, 824–829M¨ucke, A., Protheroe, R. J., Engel, R., Rachen, J. P., & Stanev, T. 2003,Astroparticle Physics, 18, 593Neshpor, Y. I., Eliseev, V. S., Jogolev, N. A., et al. 2007, Bulletin CrimeanAstrophysical Observatory, 103, 16Piron, F. 2000, PhD thesis, AA(Universit´e de Paris-Sud (Paris XI))Piron, F., Djannati-Atai, A., Punch, M., et al. 2001, A&A, 374, 895Punch, M., Akerlof, C. W., Cawley, M. F., et al. 1992, Nature, 358, 477Rebillot, P. F., Badran, H. M., Blaylock, G., et al. 2006, ApJ, 641, 740Schubnell, M. S., Akerlof, C. W., Biller, S., et al. 1996, ApJ, 460, 644Takahashi, T., Tashiro, M., Madejski, G., et al. 1996, ApJ, 470, L89 ++