First bounds on the very high energy gamma-ray emission from Arp 220

Using the Major Atmospheric Gamma Imaging Cherenkov Telescope (MAGIC), we have observed the nearest ultra-luminous infrared galaxy Arp 220 for about 15 hours. No significant signal was detected within the dedicated amount of observation time. The first upper limits to the very high energy $\gamma$-ray flux of Arp 220 are herein reported and compared with theoretical expectations.


ABSTRACT
Using the Major Atmospheric Gamma Imaging Cherenkov Telescope (MAGIC), we have observed the nearest ultra-luminous infrared galaxy Arp 220 for about 15 hours. No significant signal was detected within the dedicated amount of observation time. The first upper limits to the very high energy γ-ray flux of Arp 220 are herein reported and compared with theoretical expectations.

Introduction
The large masses of dense interstellar gas and the enhanced number densities of supernova remnants and massive young stars present in starburst galaxies suggest that they might emit γ-ray luminosities orders of magnitude greater than normal galaxies and be related to the production of cosmic rays (see, e.g., , Torres & Anchordoqui 2005. Of the taxonomy of starbursts, a place of privilege is given to luminous and ultraluminous infrared galaxies (luminous and ultra-luminous IR galaxies (LIRGs and ULIRGs, defined as having log(L IR /L ⊙ ) > 11 and 12, respectively, see Sanders & Mirabel (1996) for a review about these objects). Such star-forming environments emit a large amount of infrared (IR) radiation, because of the abundant dust molecules reprocessing of stellar UV photons. Therefore, the infrared luminosity, L IR , of a galaxy can (but not always) be an indication of star formation taking place in it. LIRGs are the dominant population of extragalactic objects in the local universe (z < 0.3) at bolometric luminosities above L > 10 11 L ⊙ , and ULIRGs are in fact the most luminous local objects. Most ULIRGs appear to be recent galaxy mergers in which much of the gas of the colliding objects has fallen into a common center (typically less than 1 kpc in extent), triggering a huge starburst (e.g., Sanders et al. 1988, Melnick & Mirabel 1990. The size of the inner regions of ULIRGs, where most of the gas is found, can be as small as a few hundreds parsecs, and populated with dense molecular environments (e.g., Gao & Solomon 2003a, 2003b) that makes them prone to have large star formation events and cosmic-ray densities. However, no LIRG, nor ULIRG, nor any other starburst galaxy has been detected in γ-rays, not even at the EGRET energy range above 100 MeV. Upper limits were imposed for M82 and NGC 253, the two nearest starbursts, as well as for many LIRGs (Torres et. al. 2004, Cillis et al. 2005. At higher energies, HESS has recently reported upper limits for NGC 253 (Aharonian et al. 2005) and we are aware of no limit reported yet for Arp 220.
Arp 220 is the nearest ULIRG (located at about 72 Mpc) and the best studied. A complete multiwavelength modeling from radio to TeV γ-rays was presented by Torres (2004), where an extensive description on the observational knowledge on this object can also be found. Arp 220 possess the record as to being the object with the highest directly measured supernova explosion rate known, with recent measurements placing it at an outrageous 4 ± 2 per year (Lonsdale et al. 2006). Such a high supernova explosion rate emphasizes the quality of Arp 220 as a possible γ-ray target.

Observations
MAGIC (see e.g., Baixeras et al. 2004, Cortina et al. 2005) for a detailed description) is a single dish Imaging Air Cherenkov Telescope. Located on the Canary Island La Palma (28.8 • N, 17.8 • W, 2200 m a.s.l.), the telescope has a 17-m diameter mirror, and it is equipped with a 576-pixel 3.5 • field-of-view photomultiplier (PMT) camera. The analogue PMT signals are transported via optical fibers to the trigger electronics and are read out by a 300 MSamples/s FADC system. MAGIC's angular resolution is approximately 0.1 • , energy resolution is about 20%, and the trigger (analysis) threshold is 55 (90) GeV.
Arp 220 celestial coordinates are: (J2000) α = 15 h 34 m 57. s 21, δ = +23 • 30 ′ 09. ′′ 5, what have allowed MAGIC to observe it always with zenith angle lower than 20 degrees. Observations proceeded from May to June, 2005, for a total time of 947 min, of which 925 min where selected -after quality checks-for further analysis. MAGIC observations were carried out in the ON-OFF mode, with a similar amount of ON and OFF data being considered for the analysis. This observation mode allows a reliable background estimation. The specific properties of the used data sample, both ON and OFF, can be found in Domingo-Santamaría (2006), where additional details on the quality checks performed and on the cut optimization using an independent Crab sample of data are also available.
The data analysis was carried out using the standard MAGIC analysis and reconstruction software (Bretz & Wagner 2003), the first step of which involves the calibration of the raw data (Gaug et al. 2005). It follows the general stream presented in (Albert et al. 2006a,b,c): After calibration, image cleaning tail cuts of 10 photoelectrons (phe) for image core pixels and 5 phe (boundary pixels) have been applied (see e.g. Fegan 1997). These tail cuts are accordingly scaled for the larger size of the outer pixels of the MAGIC camera. The camera images are parameterized by image parameters (Hillas 1985). In this analysis, the Random Forest method (see Bock et al. 2004, Breiman 2001) for a detailed description) was applied for the γ/hadron separation.
The source position-independent image parameters SIZE, WIDTH, LENGTH, CONC (Hillas 1985) and the third moment of the phe distribution along the major image axis were selected to parameterize the shower images. After the training, the Random Forest method allows to calculate for every event a parameter, the HADRONNESS, which is a measure of the probability that the event is not γ-like. The γ-sample is defined by selecting showers with a HADRONNESS below a specified value, which is optimized using a sample of Crab data which has been processed with the same analysis stream. An independent sample of Monte Carlo γ-showers was used to determine the cut efficiency. Figure 1 shows the ALPHA plots which were obtained for the whole sample of ON and OFF data, after applying the HADRONNESS cuts. The ON and OFF ALPHA distributions match reasonably well within fluctuations. No signal above the background level is observed in any of the SIZE bins. Details are given in Table 1. There, we also give the Crab rate (obtained with 1.22 hours of data under the same analysis stream applied to Arp 220) . The upper limits to the very high energy γ-ray flux from Arp 220 can then be calculated out of the number of excess and background events, in a similar way as that used in the case of LS I +61 303 (Albert et al. 2006d), using the method by Rolke and Lopez (2001). The final column of Table 1 summarizes these results.

Discussion and Concluding Remarks
The upper limits imposed to the differential flux of Arp 220 with 15.4 hours of data are above the theoretical curves at all energies. 1 All upper limits are at least about one order of magnitude above the curve obtained using the δ-function (Aharonian & Atoyan 2000) or Kamae et al. (2005) approaches for the proton-proton cross section parameterization. The latter (see the appendix in Domingo-Santamaría & Torres 2005 for a detailed discussion) are the most reliable parameterizations of the proton-proton cross section, which is devoid of the intrinsic problems of extrapolating Blattnig et al.'s (2000) formulae to high energies. The current results imply that under the same conditions of telescope performance and sensitivity of the data analysis applied, the amount of observation time needed to be devoted to Arp 220 in order to be at the level of confirming or rejecting the predictions from the γ-ray emission of this object is too high for a detector with a typical duty cycle of about 1000 hours per year, such as MAGIC. Disregarding the fact that Arp 220 is the nearest ULIRG and the galaxy with the largest supernova explosion rate we know, its distance dilutes the putative γ-ray flux it produces. MAGIC upper limits are consistent with this interpretation and with the theoretical prediction that results from a complete multiwavelength modeling 1 The theoretical curves are the result of a multiwavelength modeling discussed by Torres (2004), and . From the proton steady state population the computation of the secondary e ± proceeds considering knock-on interactions and decay of charged pions. The lepton population is let to evolve to its steady state, computing the energy losses by synchrotron emission, ionization, bremsstrahlung, inverse Compton, and adiabatic expansion, and their confinement timescale. The radio spectrum is evaluated from the steady electrons synchrotron emission, modulated by free-free absorption. Infrared emission from dust is also simulated to describe the observational data. These photons are the seed of inverse Compton process. Once the multiwavelength spectrum of the object has been reproduced, the high energy γ-ray emission is evaluated through the decay of neutral pions, and bremsstrahlung and inverse Compton of the steady electron population. Photon absorption is considered to obtain the final predictions of fluxes.   Kamae et al. (2005) approximations for the proton-proton cross section, whereas a dashed line shows the result using the parameterization proposed by Blattnig et al. (2000), extrapolated to high energies. Theoretical curves are from (Torres 2004, andTorres &. of the object. The first detection of γ-rays from starburst regions beyond our galaxy is yet to be achieved.