Table Of Content

Draftv10 pages 1–27(2014) Printed24February2015 Calibration of the Swift-UVOT ultraviolet and visible grisms 5 1 N. P. M. Kuin1⋆, W. Landsman2, A. A. Breeveld1, M. J. Page1, H. Lamoureux1, 0 2 C. James1, M. Mehdipour1, M. Still3, V. Yershov1, P. J. Brown5, M. Carter1, b K. O. Mason6, T. Kennedy1, F. Marshall7, P. W. A. Roming4,8,10, M. Siegel4, e F S.Oates1,11, P. J. Smith1, and M. De Pasquale1,9 3 2 1Mullard Space Science Laboratory/UCL, Holmbury St. Mary, Dorking, Surrey, RH5 6NT, UK 2Space Telescope Science Institure, Baltimore, MD 00000, USA ] 3NASA Ames Research Center, M/ Table S 244-40, MoffettField, CA 94035, USA M 4Department of Astronomy & Astrophysics, Penn State University,525 Davey Laboratory, UniversityPark, PA 16802, USA 5George P. and Cynthia Woods MitchellInstitute for Fundamental Physics & Astronomy, Texas A. & M. University, I . Department of Physics and Astronomy, 4242 TAMU, College Station, TX 77843, USA h 6Satellite Applications Catapult, Fermi Avenue, Harwell Oxford, Oxfordshire OX11 0QR, UK p 7NASA Goddard Space Flight Center,Code 660, MD 20771, USA - 8Space Science & Engineering Division, Southwest Research Institute, P.O. Drawer 28510, San Antonio, TX 78228-0510, USA o 9IASF Palermo, Via Ugo La Malfa 153, 90146 Palermo, Italy. r t 10The University of Texas at San Antonio, Physics & Astronomy Department, 1 UTSA Circle, San Antonio, TX 78249, USA. s 11Instituto de Astrofsica de Andaluc´ıa (IAA-CSIC), Glorieta de la Astronom´ıa s/n, E-18008, Granada, Spain. a [ 2 v Accepted: 23February2015. Received:19February2015.inoriginalform12January2015. 3 3 4 ABSTRACT 2 0 We present the calibration of the Swift UVOT grisms, of which there are two, . 1 providing low-resolution field spectroscopy in the ultraviolet and optical bands re- 0 spectively. The UV grism covers the range λ1700-5000 ˚A with a spectral resolution 5 (λ/∆λ) of 75 at λ2600 ˚A for source magnitudes of u=10-16 mag, while the visible 1 grism coversthe range λ2850-6600˚A with a spectral resolutionof 100 at λ4000 ˚A for : v source magnitudes of b=12-17 mag. This calibration extends over all detector posi- Xi tions,forallmodesusedduringoperations.Thewavelengthaccuracy(1-sigma)is9˚A in the UV grism clocked mode, 17 ˚A in the UV grism nominal mode and 22 ˚A in the r a visible grism. The range below λ2740 ˚A in the UV grism and λ5200 ˚A in the visible grism never suffers from overlapping by higher spectral orders. The flux calibration of the grisms includes a correction we developed for coincidence loss in the detector. Theerrorinthe coincidencelosscorrectionis lessthan20%.The positionofthe spectrum on the detector only affects the effective area (sensitivity) by a few percent in the nominal modes, but varies substantially in the clocked modes. The error in the effective area is from 9% in the UV grism clocked mode to 15% in the visible grism clocked mode . Key words: techniques: spectroscopy - instrumentation: spectrographs 1 INTRODUCTION instruments: the Burst Alert Telescope (BAT) to detect gamma-rays (Barthelmy et al. 2005), the X-Ray Telescope TheSwiftmission(Gehrels et al.2004)waslaunchedtopro- (XRT) to observe the X-rays (Burrows et al. 2005), and vide rapid response to gamma-ray bursts (GRB) over the the Ultraviolet and Optical Telescope (UVOT) for UV- wavelength range from gamma-rays to optical with three optical photometry and spectroscopy (Mason et al. 2004; Roming et al. 2005). UVOTspectroscopy is enabled by the inclusionoftwogrisms,theUVgrism(1700-5000˚A)andthe ⋆ email:[email protected] 2 The Swift UVOT Team visible grism (2850-6600 ˚A). These are mounted in a filter wheelwhichalsohousestheUVandvisiblelenticularfilters. The Swift grisms provide a window on the UV uni- verse to complement the high resolution HST instruments with a rapid response, low resolution option for the com- munity.The X-ray Multi-Mirror (XMM ) Optical Monitor (XMM-OM) (Mason et al. 2001) grisms provide a similar functionality but for somewhat brighter sources and with- outtherapidresponse.EarliermissionswhichprovidedUV spectroscopy include the International Ultraviolet Explorer (IUE1) (Boggess et al. 1978), and GALEX2 (Martin et al. 2003). SinceNovember2008theautomatedresponsesequence of the Swift UVOT, which governs the early exposures after a BAT GRB trigger (Roming et al. 2005), has in- cluded a 50 second UV grism exposure provided the burst is bright enough in the gamma rays. So far, this has resulted in two well-exposed UV spectra of GRB after- glows: for GRB081203A (Kuin et al. 2008) and the bright nearbyGRB130427A(Maselli et al.2014).Swifthasalsoob- VISIBLE GRISM NOMINAL MODE tainedspectraformanyotherobjects.Theseincludecomets (Bodewits et al. 2011), AGN (Mehdipouret al. 2015), su- pernovae,e.g.,Bufano et al.(2009);Brown et al.(2012)and recurrent novae (Byckling et al. 2009) where the rapid response of Swift has resulted in unprecedented early multi- VISIBLE GRISM CLOCKED MODE wavelength coverage. The UVOT uses a modified Ritchey-Chrétien optical design where light from the telescope is directed towards oneof two redundantdetectorsusing a45-degree mirror. A filterwheelallows selection of eitheraUVor optical lenticular filter, a white/clear filter, a UV grism, a visible grism, or a blocked position. Behind the filter wheel is an image intensifier configured to detect each photon event with a 2048x2048 pixel resolution. The Swift UVOT grism filters are the flight spares for theXMM-OMinstrument.Thegrismsforbothinstruments were designed using a Zemax3 optical model. The Swift UVOTinstrumentdesignandbuildprocedurewasmodified to avoid the molecular contamination which impaired the XMM-OM UV sensitivity. Therefore the sensitivity of the UVgrism ismuchbetterintheUVthanthatoftheXMM- OM grisms. The UVOT visible grism optics were blazed at 3600˚A. However, the UVOT UV grism optics were not blazed; therefore thesecond orderspectrum of thisgrism is significantandhastobeaccountedforintheanalysiswhere theorders overlap. Each of the two grisms can be operated in two modes. Figure 1. A typical detector image of the visible grism. The The so-called nominal mode is where the filter wheel is ro- top panel was observed in nominal mode, the bottom panel in clockedmode.Thezerothordersaretheshortfeatures,whilethe tated so that the grism is positioned in direct alignment longlinesarethefirstorderspectra.Intheclockedmodetheze- with the telescope optical light path. However, in order to rothorderimagesareabsentfromthetop-leftoftheimage.Note reduce the contamination by zeroth order emission of the also the change in the angle of the spectra on the detector be- background and field sources, in the so-called clocked mode tweennominalandclockedmode.Thenominalgrismimagehasa thefilterwheelisturnedsothegrismispartiallycoveredby flatbackground,whileintheclockedgrismmodethebackground thetelescope exit aperturewhich restricts thefield of view, variesacrosstheimage. blockingsomefieldstarsandreducingthebackgroundlight. Intheclocked modethefirstorderspectraof manystarsin the field of view lie in the area uncontaminated by back- 1 The IUE wavelength (2 ranges 1150-2000˚A; 1900-3200˚A) res- olutionwas0.2˚Aforhighdispersionand6˚Aforlowdispersion. ground or the zeroth order spectra of other field stars as 2 The GALEX spectral resolution (λ/∆λ) was 90for the NUV shown in Fig. 1. The clocked mode has been used exten- band(1771-2831˚A) and200fortheFUVband(1344-1786˚A). sively, though the comet and GRB observations have been 3 http://www.zemax.com done in the nominal mode. A comparison of clocked and The Swift-UVOT ultraviolet and visible grism calibration 3 Table 1.UVOTGrismSpecifications visible grism UV grism grating 300lines/mm 500lines/mm spectral resolution 100atλ4000˚A 75atλ2600˚A wavelength range(firstorder) 2850-6600˚A 1700-5000˚A wavelength accuracy(firstorder) 44˚A 17(35)f ˚A noorderoverlap(firstorder)a 2850-5200˚A 1700-2740˚A effective magnituderangeb 12-17mag 10-16mag astrometricaccuracy c,g 4′′ 3′′ scale 0.58′′/pixel 0.58′′/pixel dispersion(firstorder) 5.9˚A/pixelat4200˚A 3.1˚A/pixelatλ2600˚A fluxabovewhich20%coincidence lossd 10−14 ergscm−2 s−1 ˚A−1 10−13 ergscm−2 s−1 ˚A−1 zerothorderb-magnitudezeropointe 17.7mag 19.0mag a intheUV-grism,therangewithout2nd orderoverlapdepends ontheplacementonthedetector. b limitdepends onspectrum,seesection11.1. c offirstorderanchor point. d forlowbackgrounds. e inUVgrismfor10′′circularaperturecenteredonzerothorderaftersuccessfullycorrectingtheastrometry. f UVgrismnominalmode. g usinguvotgraspcorr inacrowdedfield. nominal images in Fig. 1 shows how effective the reduction brightness limit for coincidence loss correction, suffera fur- of zeroth order contamination is from other sources in the therlossduetointerferencefromeventsregisteredinneigh- field for those spectra falling on the left hand side of the bouring CCD pixels. image. The background due to dark current in the detector is The detector is a Microchannel plate (MCP) Intensi- very low; instead the sky background is the limiting factor fied Charge coupled device (CCD) or MIC (Fordham et al. for faint sources. The sky background in thegrisms is com- 1989). Each photon incident on the S20 multi-alkali photo- parable to that in the UVOTwhite (clear) filter since both cathodecanreleaseanelectronwhichisamplifiedamillion- grisms transmit the2800-6800 ˚A optical band. fold using a three stage MCP. The cloud of electrons hits a The sensitivity of the UVOT lenticular filter expo- P46 fast-phosphor screen, producing photons which are fed sures is decreasing slowly with time. In the UV -filters the through a fibre taper to a CCD operated in frame transfer loss is about 1% per year (Breeveld et al. 2011), while in mode. The fibre taper reduces the footprint of the image the v-band optical filter it is larger, 1.5% per year4. Most intensifier output so it fits on the exposed CCD area. The likely ageing of the MCPs (proportional to lifetime photon exposed area correspondsto 256x256 CCD pixels,but after throughput) is the main cause of the decreasing sensitivity readout the photon splash is centroided to 8 times higher which will affect all filters equally, while aging of the fil- resolution, providing an effective image that is 2048x2048 ter itself explains the different rate in the v-filter. The 1% pixels square. The nature of thecentroiding process is such sensitivity decrease is assumed to apply when the grism is that the effective size of each of the 8x8 sub-pixels on the employed as well as thelenticular filters and is used for the sky is not exactly the same, leaving a modulo-8 (MOD-8) grism calibration and data reduction. patternintheuntreatedimagewhichcanbecorrectedforin The grism image is usually stored as an accumulated dataprocessing.Howeverinformationlossthatoccurswhen image on board the spacecraft, although it is also possible more than one photon splash is registered on a CCD pixel torecordthedataasaneventlistofphotontimesandposi- withinthesameCCDreadoutinterval(coincidenceloss)can tions.Aftertransmissiontothegroundthedataisroutinely cause some pattern to remain for bright sources after cor- processedintorawimages,whicharecorrectedfortheMOD- rection. 8 pattern, followed by a correction due to small distortions inthefibretaper5 intoadetector image.Thedetectorimage As noted above, the finite time over which each exposure is integrated on the CCD, the frame time, results in is thebasis for the spectral extraction. Knowledge of the position of the spectrum on the de- coincidence losses if the photon arrival rate is high enough tector is crucial for determining the wavelength scale. The (Fordham et al. 2000). Statistically, there is a chance that UVOT spectra are formed by slitless dispersion such that multiplephotonsarrivewithinoneCCDframewithspatially the detector position depends on the position of the source overlappingpulseprofiles,inwhichcaseonlyonearrivalwill on the sky. To define the position of the spectrum on the be recorded. This means that fewer source photons are de- detector we use the position of a particular wavelength in tected than are incident on the detector, and the effect is largerwhenthereisahigherinputphotonrate,resultingin anon-linearresponsewithsourcebrightness.Makinguseof the statistical nature of the effect, the coincidence loss can 4 Updates to the UVOT calibration documents are at becorrected,andanexpressionforpointsourceshasproved http://heasarc.gsfc.nasa.gov/docs/heasarc/caldb/swift/ veryeffectiveinUVOTphotometry,e.g.,(Poole et al.2008; 5 the fibre taper distortion correction was determined using the Breeveld et al. 2010). Extremely bright sources, above the lenticularfilters 4 The Swift UVOT Team Figure2.OpticalschematicoftheUVOTUVgrismwithraysformultiplewavelengthsandorders-1(pink),0(blue),+1(green),+2(red), and+3(orange). Thecomputation was madeforanon-axis beam cominginfromthe left,whileitends onthe rightatthe entrance of theimageintensifier.ThegreenraysinthecentreareoftheUV-partofthefirstorder.Longwavelengths ofthefirstordercanbeseen interleavedwithredsecondandorangethirdorderrays. the first order, referred to as the anchor position. The an- pensator element flattens the focal plane. The UV grism is chor position in the visible grism is found at 4200˚A, in the directruledonaSuprasilsubstrate,whilethevisiblegrismis UV grism at 2600˚A. replicatedepoxy.Thegratinginthevisiblegrismisblazedto Table 1 provides a summary of the capabilities of the maximisethetransmission of thefirst order.TheUVgrism grisms. More details can be found in the main body of the grating was not blazed. Because the optics are in the con- paper. vergingbeamofthetelescope, whichuncompensatedwould The rest of this paper is organised as follows: first we induceashiftinfocus,theleadingsurfaceisslightlyconvex. provide more details on the UVOT grisms and the optical The dispersed light is refocussed before leaving the grism. model,anddiscusstheappearanceofthegrismspectrumon The grism design was optimised for the UV grism in the theimage. Wediscusshowwemap theskyposition tothat firstorderaround λ2600 ˚A,andin thevisible grism around of the anchor point of the spectrum on the image. Next we λ4200˚A. discuss the calibration of the dispersion. Then we present TheZemaxmodelwhichwasusedtodesign thegrism ourcorrectionforthecoincidencelossinthespectraandwe optics has been used to assist in the in-orbit grism cali- determine the effective areas by including the coincidence bration, but some adjustments were needed. A significant loss corrections. After discussing the second order effective reason for adjustment is that the model does not include areas, we consider the zeroth order effective area and de- the fibre taper optics between the MCP and the CCD riveaphotometricmethodandzeropointfortheUVgrism. (Rominget al. 2005), and that the glass catalogue of the Nextfollowsadescriptionofthemethodusedforextracting model does not include coefficients for the refractive ele- the spectra, and the related software. We conclude by de- ments below 2000 ˚A. The optical model was modified for scribing how to use the UVOT grism, and some additional the clocked modes with an appropriate decenter and rota- information that can be useful for the user. tion around theoptical axis of thegrism assembly6. Inthefollowing,weusethetermdefaultpositionforthe Theboresightofthemodelwasalignedtotheobserved spectrum placement in the middle of the detector, roughly boresight for all modes, but thenafurtheradjustment of (- at the boresight of the instrument. This is the normal op- 60,0)pixelswasneededfortheUVgrismtoalignthemodel eratingposition forsources;otherpositionsareatanoffset. dropinfluxintheleft topcornerofthedetectorduetothe Thetermanchorisusedtodesignatethepositionthatfixes clockingtomatchtheobservationsfromcalibrationspectra. thewavelengthscale,inthatthedispersionismeasuredrela- The optical model predicts that the point spread tivetothatposition.BymodelwemeantheZemaxoptical function (PSF, which describes the distribution of where model, eithercorrected or not. monochromatic photons fall on the detector) increases in size in the UV grism towards the red. This is illustrated in Fig. 3 where the 2-dimensional model PSF has been integrated normal tothedispersion, illustrating thePSFvaria- 2 DESCRIPTION OF THE GRISMS AND THE tion as a function of wavelength. OPTICAL MODEL The large width of the PSF in the UV grism for wavelengths longer than 4500 ˚A causes the spectrum to appear The“grisms” areactually madeupoftwooptical elements: agrismandatiltcompensatorasshowninFig.2.Thegrism provides on-axis dispersion by means of a prism and trans- mission grating, but the focal plane is tilted. The tilt com- 6 Therotationiscalleda”tilt”intheZemaxmodel. The Swift-UVOT ultraviolet and visible grism calibration 5 1.0 wavelength 7000.0 6600.0 6200.0 0.8 5800.0 5400.0 d) 5000.0 cke0.6 4600.0 o 4200.0 V cl 3800.0 U 3500.0 F (0.4 3200.0 PS 2850.0 2600.0 2350.0 0.2 2100.0 0.0 2000 3000 4000 5000 6000 7000 λ(Å) Figure 4. A section of a ground calibration image of a narrow Figure 3. The model point spread function of the UV grism is bandfilterexposurearound260nmintheUVgrismoutlinesthe illustratedasafunctionofwavelength. positionsoftheorders.Acartoonofthetypicalrelativepositions ofthespectralordersisshownabovethedata.Thefirstorderin blue, the second order in red, the third order in green, and the smoothed out with wavelength at thesewavelengths.In the zerothorderinbrown.Thecurvatureofthespectralorders,which UV, despite the small PSF width predicted by the model, onlyoccursintheUVgrismhasbeenexaggeratedforillustration. otherfactorscontributetobroadeningthespectrum.Specif- ically, the actual PSF below 3000 ˚A is thought to be dominated by the transverse spreading of the electrons leaving the cathode. Spacecraft jitter during the observation is re- movedusingtheshiftandaddmethod(Roming et al.2005). In both grisms, the zeroth order extends over several This results in a FWHM of the PSF of typically 3 pixels, pixels,andthehigherordersoverlap.However,inthevisible about10˚A,intheUVpartofthespectrumintheUVgrism grism the zeroth order brightness and second order bright- and about 20 ˚A in thevisible grism. ness are much less than the first order. As a result, the Fig.2displaysaUVgrismmodelcalculationforanon- contamination of the first order spectrum from higher or- axis source of the rays for several UV wavelengths and for der light is small, and contamination by zeroth orders of the five orders that can be registered on the detector. The field stars is usually small. The spectral tracks in the vis- first, second and third orders overlap, while the zeroth and ible grism are straight, which means the spectrum is easy minus-first orders are separate. to extract. The angle of the spectra in the detector frame The model zeroth order spectrum is dispersed in a hy- varies slightly over the detector, however. Example images perbolic fashion, as discussed in more detail in Section 9.1. are shown in Fig. 1. As a result photons with wavelengths longer than about The zeroth and second order in the UV grism are of λ3500 ˚A fall within a single pixel and in the UV grism the comparable brightness to the first order. Like in the visible UV spectrum forms a very weak tail that extends for 200 grismthezerothorderisextended,withaverysmalltaildue pixels, see for example theinset in Fig. 5. totheUVresponse, and thefirst and higher orderstend to The dispersion angle, or slope of thespectra in thede- overlap.UnlikethevisiblegrismtheUVpartofthespectrum tector image, varies by about 5 degrees over the detector in each order is generally curved. The curvature is largest for a given mode. The angle near the centre of the image in the upper right and lower left corners of the detector is different for each mode due to the positioning of the imageandgoesintheoppositedirection.Thespectraltrack grism in the filter wheel and filter wheel clocking, and is is straight near the centre of the image, where there is full 144.5◦ (UV grism clocked mode), 151.4◦ (UV grism nomi- overlap of first and second orders. Depending on where the nal mode), 140.5◦ (visible grism clocked mode), and 148.1◦ spectrumfallsonthedetector,thesecondorderoverlapcan (visible grism nominal mode). startinthefirstorderassoonas2740˚Aoraslateas4500˚A. Themodelpredictsthevariationofthedispersionangle A simplified drawing of the layout of the UV grism image overthedetector,butdoesnot includethecurvatureof the can befound in Fig. 4. spectra in the UV grism. The predicted model dispersion Theappearanceoftheobservedcurvatureanddisplace- angle is used in thespectral extraction; see Section 10. mentintheUVgrismisevidentinimageswithaverybright source,e.g.,Fig.13.Themagnitudeoftheeffectisafunction ofdetectorposition,beingverysmallatthedefaultposition. Adoptingastraight lineinthedispersion directionasrefer- 3 THE APPEARANCE OF THE SPECTRUM ence, the maximum curvature offset in the cross-dispersion ON THE DETECTOR direction varies from about 16 pixels in the lower left up to The appearance of the grism spectrum differs between the minus25 pixelsin thetop right hand detectorcorner. UVand visible grisms. Thisapplies equally to nominal and In the visible grism there is no noticeable curvature or clocked modes in thesame grism. anyoffset ofthehigherordersatanypointon thedetector. 6 The Swift UVOT Team 3.1 Bright sources One of the characteristics of the grism images is that there is a MOD-8 pattern of dark (low count rate) pixels next to the spectra of bright sources; see for example the nominal mode UV grism image of the region around WR52 in Fig. 5, with the highlighted WR52 spectrum in the centre. Thecross-hatchedMOD-8patternisasignthatcoincidence loss is present in the spectrum. Even when it is present, a correction for thecoincidence loss is often still possible. Forsources with ab magnitudebrighterthan about 17 thezerothorderintheUVgrismdevelopsadarkpatterned region because of coincidence loss, and when brighter than 13th magnitudes,aregionwitha49pixelradiusaroundthe source is affected and can cause part of a nearby spectrum to beunreliable. Occasionally, very bright stars (V < 8th mag) are in a grism image. These can cause problems because their readout streak (caused by exposure of the column during the image transfer to the CCD readout area) leads to columns ofelevatedcountsacrosstheimage, e.g.,Page et al.(2013). Figure 5.Atypical detector imageofthe UVgrisminnominal When the readout streak crosses a spectrum, it does so at mode. The top inset highlights the extended disconnected UV - anangleandbackgroundsubtractionandthecorrection for tailinthezerothorderoftheUV-brightsourceinthecentre.The coincidence loss may beaffected. disconnection is due to a combination of detector effective area In parts of calibration spectra brighter than the and interstellar extinction. The background around the bright coincidence-loss upperlimit of5countsperframe asmaller zeroth order has been eaten away due to coincidence loss, and countrateisobservedthantheexpectedcountrateofnearly shows up as white pixels. The bottom inset shows the MOD-8 1. This is thought to be caused by the amplified photon patternaroundabrightspectrum.Noticethemanyzerothorders splashsaturatingneighbouringpixels.Thiswouldcausethe fromweaker sources. centroidingtofailinwhichcasethoseeventsarenotcounted. However,atsuchbrightnessthecoincidencelossdistortsthe ically could be used to do the mapping from sky to anchor source spectrumseverely so confusion with thespectrum of positionaftercorrectingfortheimagescalingcausedbythe a fainter source is unlikely. fibre taper. However the model does not predict this dis- Other features due to scattered light that are seen in tortioncompletelyaccurately.Somedifferencesremainfrom the lenticular filter images (Breeveld et al. 2010) may also theobservedanchorpositionswhichareconsideredlikelyto be present in grism images. Experience shows that they do be due to the unknown overcorrection due to thelenticular notoccurfrequentlyenoughtoaffectthegrismspectroscopy filters. in practice. Thedistortioninthedetectorimageofthezerothorder positions due to the grism optics was calibrated using catalog positions of the USNO-B1 catalog for several fields, and 3.2 The fibre taper distortion is made available7 in theSwiftCALDB8. Images taken through the UVOT lenticular filters are spa- tiallydistortedbytheimageintensifierandfibretaper,with asmallcontributionfromthelenticularfilteritself.Thecor- 3.3 The cross-dispersion profile rection has been determined to be the same for all lenticu- The final footprint of the light entering the detector is pri- larfilters,soasingle distortion correction isneededtomap marily broadened by transverse diffusion of the electron positions on the detector to those on the sky. Using this cloud in the gap between the cathode and MCP, which has distortion map, the standard ground processing produces a a profile similar to a Gaussian. The grism optics also add corrected image called thedetector image. broadening so that the spectral profile is of different width The grism causes further distortion and this might be for the different orders. Finally, coincidence loss further af- wavelength and order dependant. fects the profile. In order todo theanchorpoint calibration, we need to IntheUVOTPY software (seeSection 10)a fitismadeof correctforthedistortionoftheanchorpoints,sowecanfind the count rate as a function of pixel distance to the centre amappingfromtheskypositiontotheanchorposition.We of thetrack using a gaussian distribution also need to map the zeroth order positions so we can get an astrometric solution for thegrism image. f(x)=a e−(x−σxo)2 (1) Thesamedistortionmapasusedforthelenticularfilters isused toconverttherawgrism image toadetectorimage. with x, the cross-dispersion pixel coordinate, a the peak Ittakesoutthemajordistortionduetotheimageintensifier and fibretaper, though it may over-correct somewhat since it also includes thelenticular filter part of thecorrection. 7 swugrdist20041120v001.fits The model predicts the grism distortion, so it theoret- 8 http://heasarc.gsfc.nasa.gov/docs/heasarc/caldb The Swift-UVOT ultraviolet and visible grism calibration 7 optical model uses the optical set-up to predict the dispersion,orderoverlap,PSF,andthroughput,andpredictstheir 1.0 variation as a function of the source position in the field of view.Themodelcontainsamajorpartofthephysicsofthe opticsandthusitconstrainsthecalibration,providesaway 0.8 typical error uv to verify observed parameters, and allows us to extend the on typical error visible calibration to all parts of the detector. As a result we can acti0.6 have a more reliable calibration by determining corrections e fr in the form of alignments and by scaling the model where at nt r appropriate. ou0.4 c UV grism clocked 4.1 Method of implementation 0.2 UV grism nominal visible grism clocked FortheUVwavelengthcalibrationsources,weselectedWolf- visible grism nominal Rayet stars with a fairly good coverage of bright emission 0.00 2 4 6 8 10 12 linesintheUV;seeTable2.Thesesourcesareontheupper distance from central peak (pixels) range of brightness that can be observed with the UVOT grisms. Forthecalibration of effectivearea and coincidence Figure 6.Thecross-dispersioncountrateswithinaslitasfunc- loss we used mainly sources with reference spectra in the tionofthepixeldistancetothecentreofthespectraltracklevel CALSPEC9 database which are flux calibrated to typically off around 7.5 pixels in the UV grism, and around 5.5 pixels in 2-3%. the visible grism. The profiles shown are for weak spectra with We selected anchor positions for this calibration at thelowestpossiblecoincidenceloss(WD1657+343). 2600˚AfortheUVgrismand4200˚Aforthevisiblegrismin thefirst order. Therefore, theanchorposition on thedetec- countrate,xothecentreofthespectraltrack,andσcontrols torissimilartothepositionofthesourceintherawimagein thewidth of the gaussian fit. a lenticular filter when taken alongside thegrism exposure. In the first order UV grism, σ is about 2.9 pixels at The curvaturewas measured relative to the model dis- 1700 ˚A, growing to about 3.3 pixels at 6000˚A. The second persionangleforthefirstandsecondordersforallUVgrism order is broader, with σ ≈ 4.5 pixels wide. The values are calibration spectra and a correction was derived, expressed slightly smaller for thevisible grism, at 2.7 pixels. in terms of polynomials. The polynomial coefficients vary Measurements were made of the profile normal to the with theanchorposition of thespectrum butthecurvature dispersion, which we will call the cross-track profile. The isalways thesame in thesame grism modeat thesame de- measurements were made in the region of no order overlap tectorposition.Bisplineswerefittedtothepolynomialcoef- byrepeatedlyextractingthespectrawithvaryingextraction ficients,whichallowtheretrievalofthecurvatureatanyan- widths. The cross-track profile is not completely gaussian, chorpointonthedetector.Thecoefficientsofthecurvature but falls off more steeply in the wings. The plot of the en- calibration have been implemented in the UVOTPY software closed count rates, similar to encircled energy in a point (Kuin2014),seeSection10.Thespectrainthevisiblegrism source but for a linear feature (see Fig. 6) shows the cu- are not measurably curved,but straight. mulativedistribution starting from thecentreas afunction Thevariationofthewidthofthefirstand(wherepossi- of the pixel distance from the centre to the border of the ble) second spectral order was determined in the UV grism extraction “slit”. The width of the spectral track is seen to and also compared from image to image. Because of width besmallerforthevisiblegrismthanfortheUVgrism.This variations along the spectrum, the spectral extraction was profile can be used as an aperture correction, see Section designed to keep the same enclosed energy for a consistent 11.2. fluxcalibrationbyadjustingforslowvariationsinthewidth Thewidthofthespectraltrackchangeswithincreasing of thespectrum duringspectral extraction. coincidencelossduetothedevelopingMOD-8pattern.This Once we understood broadly the geometry, for each variation introduces an uncertainty larger than 20% in the grism mode the analysis of anchor position, dispersion, co- aperture correction when the coincidence loss is more than incidence loss and effective area was repeated as discussed 20%. Therefore, a smaller aperture for the spectral extrac- in the nextsections. tion with aperture corrections should only beused for faint spectra. 5 CALIBRATION OF THE ANCHOR POINT 5.1 The astrometric correction of the grism image 4 CALIBRATION APPROACH Theaspectisinitiallycorrectedusingthebestattitudefrom Beforeafullcalibrationofthewavelengthandfluxcouldbe ′′ thespacecraftwhichisaccuratetypicallytowithin1.3 (1σ); attempted, we needed to have a good aspect correction for see Breeveld et al. (2010). the grism images so we know the sky location of the bore- The positions of weak zeroth orders can be used to de- sight, and measure the width and curvature of the spectra rive an improved aspect solution. The aspect correction is overthe detector. These basic calibrations were done first. Wehaveadoptedanapproachwhichmergesthecalibra- tionobservationswiththegrismZemaxopticalmodel.The 9 http://www.stsci.edu/hst/observatory/cdbs/calspec.html 8 The Swift UVOT Team for the visible grism, no IUE spectrum is available, but for the same spectral type the IUE spectrum of WR103 provides a similar spectrum with the line identifications from Niedzielski and Rochowicz (1994) and a good ground based spectrum from Torres and Massey (1987). For WR4 nogroundbasedspectrumwasused,butthelineswereeas- ilyidentifiedfrom comparison totheotherspectra.Though WR86 is a binary, the spectrum is dominated by the WC spectrum;radialvelocitiesmayleadtoshiftsof<2˚Awhich are negligible at theresolution of thegrisms. For the wavelength verification some spectra from the flux calibration sources were used. The lines used in those spectra are mainly Mg II 2800˚Aand theHydrogen lines. 5.3 First order anchor position - fitting to model Inordertocalibratetheanchorposition,calibrationspectra accompaniedbyanimageinalenticularfilterweretakenof WRstars, seeTable2,whilethepointingwasoffset sothat thespectracoveredthedetector;seethetoppanelsinFigs. 8, 10, 11, and 12 where the locations of the anchor points on the detector are plotted. Although the emission lines in these stars are broad, their width is not an impediment as it is close to thespectral resolution of theinstrument. Figure 7. A typical detector image of the UV grism in clocked Thecalibrationspectrawereobservedinaspecialmode, mode. wherealenticularfilterexposureistakenjustbeforeand/or after the grism exposure. The spacecraft pointing is not donebyapplyingtheuvotgraspcorrftool10 whichusesthe changed during the sequence although there may still be appropriatedistortion filefromtheSwiftCALDBdescribed some drift in the pointing between the exposures. Within previously in Section 3.2. The aspect corrected coordinates an exposure,thepositions arecorrected usingtheon-board are written to the FITS header as the WCS-S world coordi- shift-and-addalgorithm (Poole et al. 2008).Theposition of natesystemkeywordsincludingthekeywordsfortheSimple thetargetinthelenticularimagescanthenbecorrelatedto Imaging Polynomial (SIP) convention (Shupeet al. 2005) that of theobserved anchor in thegrism. which capturethezeroth order anchor distortion. The position of the source relative to the boresight in The accuracy of the astrometric correction dependson thelenticular image and theanchor position relative to the the success of the uvotgraspcorr program. In the visible boresight in the grism image are related in a fixed manner. grism, with weaker zeroth orders due to the blazing, the Ignoring thedistortion, theconversion from lenticular filter errors tend to be larger. These are reported in terms of the to grism position is a shift in detector X,Y position and a anchorpointaccuracyinTable3,andinthemiddlepanelof scalefactor.Norotationisneccessary,duetothecoordinates Figs.8,10,11,and12.Theresultsshowascatterthatvaries being tied to thedetector orientation. mainly by target while for a given target different images For each observed spectrum, bright spectral lines were tend to have similar errors. Tests show that the program identifiedintheimageclosetotheanchorpoint,whereafter fails in 3-4% of the fields, in which case the correction can the anchor point on the detector image was determined for still bedone byhand. eachspectrum.Theanchorpositionsforthedefaultposition (with thesource at theboresight) havebeen given in Table 3 in detector coordinates11, and are shown in the top panel 5.2 Reference data in Figs. 8, 10,11, and 12 as a blue cross. The anchor and wavelength dispersion calibration consists A comparison of the observed grism anchor positions of the determination of the scaling of the model by using andsourcepositionsinthecorrespondinglenticularfilter(s) calibrated spectra of bright Wolf-Rayet (WR) stars. implies a pixel scale in the grism image of 0.58±0.04 arc- Our WR stars were observed by IUE, and have also sec/pixel,largerthanthe0.502valueforthelenticularfilter, ground based spectra available with sufficiently good wave- though it should be noted that thepixelscale varies dueto length calibration to determine the wavelengths of spectral distortion. emissionlines.MajorlinesusedintheWC-typespectraare: Inthelenticularimagethepositionofthesourcecanbe SiII1816˚A,CIII1909˚A,2297˚A,CIV2405,2530,2906˚A, found from the sky position. Given the source sky position O IV 3409 ˚A, C III 4069, 4649 ˚A, C IV 5801 ˚A, while andtheFITSWCSheaderintheaspectcorrectedlenticular for the WN-type spectra we used: He II 2511, 2733, 3203, filter image, we derive the astrometrically corrected source 4686 ˚A. Further minor emission lines, sometimes blended, are present in our spectra. For WR121, which was used 11 Thedetector coordinatesareconvertedherefrommmtopix- els by a centre of [1100.5,1100.5], and a scale factor of 0.009075 10 http://heasarc.gsfc.nasa.gov/docs/software/ftools/ mm/pix The Swift-UVOT ultraviolet and visible grism calibration 9 Table 2.Calibrationtargetsused. name/ID sp. J2000 position used reference type RA DEC for spectrum,notes WR1 WN4 00:43:28.4 +64:45:35.4 1 IUE,* WR4 WC5+? 02:41:11.7 +56:43:49.7 1 IUE WR52 WC4 13:18:28.0 -58:08:13.6 1 IUE,# WR86 WC7(+B0III-I) 17:18:23.1 -34:24:30.6 1 IUE,# WR121 WC9d 18:44:13.2 -03:47:57.8 2 IUE,$ WD0320-539 DA 03:22:14.8 -53:45:16.5 3,4,5 CALSPEC WD1057+719 DA1 11:00:34.2 +71:38:03.9 3,4,5 CALSPEC WD1657+343 DA1 16:58:51.1 +34:18:53.5 3,4,5 CALSPEC GD153 DA1 12:57:02.3 +22:01:52.7 5 CALSPEC GSPCP177-D F0V 15:59:13.6 +47:36:41.9 3,4,5 CALSPEC GSPCP41-C F0V 14:51:58.0 +71:43:17.4 3,4,5 CALSPEC BPM16274 DA 00:50:03.7 -52:08:15.6 4,5 ESOHSTstandards GD108 sdB 10:00:47.3 -07:33:31.0 4,5 CALSPEC GD50 DA2 03:48:50.2 -00:58:32.0 4,5 CALSPEC LTT9491 DB3 23:19:35.4 -17:05:28.5 4,5 CALSPEC WD1121+145 sdB 11:24:15.9 +14:13:49.0 3,4,5 CALSPEC G63-26 sdF 13:24:30.6 +20:27:22.1 3,4,5 STIS-NGSLv2 AGK+81266 DB2 09:21:19.2 +81:43:27.6 5 CALSPEC BD+254655 DB0 15:51:59.9 +32:56:54.3 5 CALSPEC BD+332642 B2IVp 15:51:59.9 +32:56:54.3 5 CALSPEC use: 1:UVgrismanchorandwavelength calibration 2:visiblegrismanchor andwavelength calibration 3:UVgrismfluxcalibration 4:visiblegrismfluxcalibration 5:coincidencelosscalibration CALSPEC TheHubbleSpaceTelescopecalibrationspectradatabaseatSTScI IUE ESAVilspaarchivefortheInternational UltravioletExplorer ∗ Hamannetal.(1995) $ TorresandMassey(1987),usedWR103toIDlines # CDS catalogIII/143Torres-DodgenandMassey(1988) spectraltypesfromvanderHucht(2001),CookeandSion(1999),CALSPEC. STIS-NGSLv2 http://archive.stsci.edu/prepds/stisngsl/ ESOHSTstandards http://www.eso.org/sci/observing/tools/standards/spectra/hststandards.html Table 3.Defaultanchorpositionsandwavelength accuracy. Grismmode anchor1 anchor2σ wavelength anchor 2σ wavelength default accuracy(˚A) accuracy(˚A)4 accuracy(˚A) accuracy(˚A)4 position detector centre2 detector centre2,3 fulldetector fulldetector anchor positiondeterminedusingamodecombinedwithlenticularfilter UVnominal [1005.5,1079.7] 30 7,18,36 35 8,16,34 UVclocked [1129.1,1022.3] 12 8,11,21 17 7,22,18 visiblenominal [1046.3,1098.3] 30 5,10,6 44 6,13,6 visibleclocked [1140.7,1029.6] 48 5,14,13 44 4,13,12 anchorpositiondeterminedusingastrometryfromuvotgraspcorr UVnominal [1005.5,1079.7] 53 46,15,22 53 51,17,25 UVclocked [1129.1,1022.3] 47 8,11,21 47 7,12,18 visiblenominal [1046.3,1098.3] 88 3,10,8 88 5,13,7 visibleclocked [1140.7,1029.6] 118 9,16,14 118 8,16,12 1 firstorder,indetector coordinates 2 Thedetector centreisdefinedbyimagepixelsbetween 500and1500inXandY. 3 2σ errorsforthreerangesintheUVgrismofλ<2000˚A,2000<λ<4500˚A,4500˚A<λ, andinthevisiblegrismofλ<3100˚A,3100<λ<5500˚A,5500˚A<λ. 4 excludingerrorsduetotheanchor. 10 The Swift UVOT Team uv grism nominal mode 2000 0 observed WR86 model nates1500 WR52 d (pixels)−−10500 di ∆ or −150 o c1000 or 2000 3000 4000 5000 6000 ct 30 dete 500 2205 mobesaenrv eerdro-mrodel Y- 15 )Å 10 location on detector by anchor position ∆λ( 5 0 0 0 500 1000 1500 2000 −5 X-detector coordinates in pixels −10 −15 2000 3000 4000 5000 6000 λ(Å) 0.025 without lenticular P177D filter 0.020 GD153 Figure9.Exampleofthedeterminationofthewavelengthaccu- 0.015 P041C racyforonecalibrationspectrumintheclockedUVgrism.Inthe 0.010 G63-26 upperpanel∆disderivedfromtheanchordistanceinpixels,after 0.005 subtractingalinearconstantdispersionfactor,andillustratesthe 0.000 higher order variation of the dispersion. The lower panel shows WR52 theerrorsinthemeasuredwavelengths. 0.030 0.025 with lenticular WR86 filter 0.020 P177D 0.015 P041C model broughtall themodel and observed anchorpositions 0.010 G63-26 to within 16 pixels. The position differences were not dis- 0.005 placedrandomly,soabisplinefittotheXandYcoordinate 0.000 −60 −40 −20 0 20 40 60 differencesprovidedafinalcorrection. Theorderofthebis- anchor offsets (in Å) plines was kept as small as possible in order to keep the numberof parameters low13. The corrected model was used to tabulate a lookup 0.12 table of anchor positions on a grid of field positions. The 0.09 lookup table was used subsequently to rederive the anchor 0.06 positions to obtain an estimate of the accuracy. 0.03 Any inaccuracy of the anchor position leads to a shift 0.00 inthewavelengthscale.Sincemanycalibrationobservations 0 5 10 15 20 had a lenticular filter before and after the grism exposure, standard deviation ∆λ(Å) we also obtained an estimate of the pointing drift during theexposures (often 1 kslong) of typically 6 pixels (≈ 3′′), thoughalargerexcursionbetweenexposures(10pixels)was Figure 8. The anchor and wavelength calibration for the UV observedinasingleobservation.Thisisconsistent with the grism nominal mode. The top panel shows the positions on the accuracy seen in the anchor positioning. In the UV grism detector for each spectrum used in the wavelength calibration. nominal mode calibration there was only one lenticular fil- The position of the anchor for a spectrum at boresight is indi- catedwithabluecross.Thesecondpanelshowsthemeasurement terin theobservation sonocorrection fordrift between the of the wavelength shift due to errors in the anchor position for grism and lenticular filter exposure was possible. This ex- twomethods,byusinguvotgraspcorr,andwithalenticularfilter plains the larger errors in the anchor position calibration alternatively. The third panel shows the histogram of the stan- (thusthewavelengths were shifted more). dard deviation of the errors in the measured wavelengths after Calibration observations taken before 2008 did not in- removingtheanchor error. clude a lenticular filter. This includes most of the flux calibration spectra for the visible grism. When using the sky positionfromtheimageheadertodeterminetheanchorpo- position on the lenticular filter image and thus the source sition without using a lenticular filter image taken next to positionrelativetotheboresightposition.Thatrelativepo- sition is converted into the field coordinate12 for input to the grism image, the anchor error is due to the accuracy of theuvotgraspcorrprogramasdiscussedinSection5.1above. the model. We now can use the model to find the anchor Verification withindependentdatawas doneusingflux position on the grism detector image, provided the model has been properly scaled. We found that a simple scaling of the pixel size in the 13 e.g.,fortheUVgrismclocked modetherewere24parameters used tofit 50 positional data; other grismmodes used fewer pa- 12 the field coordinate is the angular coordinate relative to the rameters to fit a comparable number of data points. Therefore, boresight enoughfreeparametersremain.