Microlensing variability in the gravitationally lensed quasar Q2237+0305 = the Einstein Cross, I. Spectrophotometric monitoring with the VLT
aa r X i v : . [ a s t r o - ph ] J a n Astronomy&Astrophysicsmanuscript no. 8703 c (cid:13)
ESO 2018October 30, 2018
Microlensing variability in the gravitationally lensed quasarQSO 2237 + ≡ the Einstein Cross ⋆ I. Spectrophotometric monitoring with the VLT
A. Eigenbrod , F. Courbin , D. Sluse , G. Meylan , and E. Agol Laboratoire d’Astrophysique, Ecole Polytechnique F´ed´erale de Lausanne (EPFL), Observatoire de Sauverny, 1290 Versoix, Switzerland Astronomy Department, University of Washington, Box 351580, Seattle, WA 98195, USAReceived ... ; accepted ...
ABSTRACT
We present the results of the first long-term (2.2 years) spectroscopic monitoring of a gravitationally lensed quasar, namely the Einstein CrossQSO 2237 + / FORS1 spectra to accurately separate the spectrum of the lensing galaxy from the spectra of the quasarimages. Accurate cross-calibration of the 58 observations at 31-epoch from October 2004 to December 2006 is carried out with non-variableforeground stars observed simultaneously with the quasar. The quasar spectra are further decomposed into a continuum component and severalbroad emission lines to infer the variations of these spectral components.We find prominent microlensing events in the quasar images A and B, while images C and D are almost quiescent on a timescale of a fewmonths. The strongest variations are observed in the continuum of image A. Their amplitude is larger in the blue (0.7 mag) than in the red(0.5 mag), consistent with microlensing of an accretion disk. Variations in the intensity and profile of the broad emission lines are also reported,most prominently in the wings of the C III] and center of the C IV emission lines. During a strong microlensing episode observed in June2006 in quasar image A, the broad component of the C III] is more highly magnified than the narrow component. In addition, the emissionlines with higher ionization potentials are more magnified than the lines with lower ionization potentials, consistent with the results obtainedwith reverberation mapping. Finally, we find that the V-band di ff erential extinction by the lens, between the quasar images, is in the range0 . − . Key words.
Gravitational lensing: quasar, microlensing — Quasars: general. Quasars: individual QSO 2237 +
1. Introduction
The gravitational lens QSO 2237 + z s = .
695 quasar gravita-tionally lensed into four images arranged in a crosslike patternaround the nucleus of a z l = . . ′′ .A few years after this discovery, Schneider et al. (1988)and Kent & Falco (1988) computed the first simple models ofthe system, leading to the conclusion that this system was verypromising to study microlensing. Indeed, the predicted time de- ⋆ Based on observations made with the ESO-VLT Unit Telescope lays between the four quasar images are of the order of a day(Rix et al. 1992, Wambsganss & Paczy´nski 1994), meaning thatintrinsic variability of the quasar can easily be distinguishedfrom microlensing events. In addition, the particularly smallredshift of the lensing galaxy implies large tangential veloci-ties for the microlenses. Furthermore the quasar images formright in the bulge of the lens where the stellar density is thehighest. The combination of these properties makes microlens-ing events very likely in the Einstein Cross and very rapid, withtimescales of a few weeks to a few months. Indeed, Irwin et al.(1989) reported significant brightness variations of the bright-est quasar image A, which they interpreted as the first detectionever of microlensing in the images of a multiply-imaged quasar.Since then, microlensing events have been observed in sev-eral other gravitationally lensed quasars, and are expected tooccur in virtually any quadruply lensed quasar (Witt et al.1995). Probably the most compelling examples of microlens-ing light curves are given by the Optical Gravitational Lensing
A. Eigenbrod et al.: Microlensing variability in the Einstein Cross
Experiment (OGLE) (Wo´zniak et al. 2000a, Udalski et al.2006). Started in 1997, this project monitors regularly thefour quasar images of QSO 2237 + ffi cult to disentangle variationsin the continuum from variations in the broad emission lines(BELs). Both types of regions are a ff ected by microlensing,but in di ff erent ways depending on their size.Microlensing of an extended source can occur when its sizeis smaller than or comparable to the Einstein radius of a star,i.e. of the order of 10 cm or 10 − pc in the case of the EinsteinCross (Nemiro ff cm or1 pc, hence leaving little room for BEL microlensing. However,more recent reverberation mapping studies revise this down-wards, to 10 cm (Wandel et al. 1999, Kaspi et al. 2000),which is also consistent with the disk-wind model of Murrayet al. (1995). Inspired by these numbers, Abajas et al. (2002)and Lewis & Ibata (2004) investigated BEL microlensing infurther detail and computed possible line profile variations forvarious BLR models.Observations of significant continuum and BEL mi-crolensing have been reported in a number of systems(QSO 2237 + − − + + − − + − Fig. 1.
VLT / FORS1 field of view showing the lensed quasarQSO 2237 + ff ect of sky transparency.
2. Observations
We acquired our observations with the FOcal Reducer and lowdispersion Spectrograph (FORS1), mounted on Kueyen, theUnit Telescope + . ′′ per pixel. With this resolution, we observed a maximum of 8objects simultaneously over a field of view of 3 . ′ × . ′ . Oneslit was aligned along two of the quasar images and four slitswere centered on foreground PSF stars. We placed the remain-ing slits on empty sky regions and used them to carry out skysubtraction of the quasar data. . Eigenbrod et al.: Microlensing variability in the Einstein Cross 3
1" E NC DBAmask 2mask 1
Fig. 2.
FORS1 R-band acquisition image of QSO 2237 + − − . ′′ . The Position Angle (PA)of mask 1 is PA = + . ◦ and that of mask 2 is PA = + . ◦ .Two masks were designed to observe the two pairs ofquasar images. The PSF stars in both masks were the same.Fig. 2 shows the slit positioning with respect to our target. Thefirst mask was aligned on quasar images A and D, while thesecond was aligned on images B and C. The masks were ro-tated to position angles that avoid clipping of any quasar im-age. This is mandatory to carry out spatial deconvolution of thespectra.Our observing sequences consisted of a short acquisitionimage, an “image-through-slit” check, followed by a consec-utive deep spectroscopic exposure. All individual exposureswere 1620 s long. We list the journal of our observations inTable 1. The mean seeing during the three observing seasonswas 0 . ′′ . We chose a slit width of 0 . ′′ , approximately match-ing the seeing and much smaller than the mean separation of1 . ′′ between the quasar images. This is mandatory to avoidcontamination of an image by the others.We used the G300V grism in combination with the GG375order sorting filter. For our slit width, the spectral resolutionwas ∆ λ =
15 Å, as measured from the FWHM of night-skyemission lines, and the resolving power was R = λ/ ∆ λ ≃ λ = < λ < .
69 Å per pixelin the spectral direction. This configuration favors spectral cov-erage rather than spectral resolution, allowing us to follow thecontinuum over a broad spectral range, starting with the veryblue portion of the optical spectrum. Even so and in spite of R = Table 1.
Journal of the observations taken on 31 epochs.
ID Civil Date HJD Mask Seeing [ ′′ ] Airmass1 13 − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − Fig. 3.
Spectra of the four PSF stars. The spectra in each panelcorrespond to di ff erent observing epochs, chosen to span thefull length of the monitoring. The IDs of the observing epochs,as given in Table 1, are indicated. The di ff erences in flux aremainly due to the presence of thin clouds. The purpose of us-ing these stars as flux cross-calibrators is precisely to eliminatethese di ff erences, both in intensity and shape.Finally, it is worth emphasizing that all our VLT data usedin the present paper were obtained in service mode, withoutwhich this project would have been impossible.
3. Data analysis
The data reduction followed the same procedure described indetail in Eigenbrod et al. (2006b). We carried out the standardbias subtraction and flat field correction of the spectra usingIRAF . We obtained the wavelength calibration from the spec-trum of helium-argon lamps. All spectra, for the object and forthe PSF stars were calibrated in two dimensions.Only one single exposure was taken per mask and perepoch. For this reason, the usual cosmic-ray rejection schemeapplied to multiple images could not be applied. Instead, weused the L. A. Cosmic algorithm (van Dokkum 2001), thatcan handle single images. We visually inspected the cosmic-ray corrected images to check that no data pixel was a ff ectedby the process, especially in the emission lines and in the datawith the best seeing. IRAF is distributed by the National Optical AstronomyObservatories, which are operated by the Association of Universitiesfor Research in Astronomy, Inc., under cooperative agreement withthe National Science Foundation.
Fig. 4.
Flux correction for di ff erent epochs with respect to thereference epoch − − ff erent spectral ranges are due to di ff erentclippings of the spectra by the edges of the CCD.We removed the sky background in a di ff erent way in thespectra of the PSF stars and in those of the gravitational lens.For the PSF stars, which are small compared with the slit length(19 ′′ ), we used the IRAF task background . This task fits asecond order Chebyshev polynomial in the spatial direction tothe areas of the spectrum that are not illuminated by the ob-ject, and subtracts it from the data. As the lensing galaxy inQSO 2237 + + Once the cosmic rays and sky background were removed, weapplied a flux cross-calibration of the spectra as described byEigenbrod et al. (2006a), using the four PSF stars. The spectraof these stars are shown in Fig. 3 for five di ff erent observingepochs. Our observations show that these stars are non variable. . Eigenbrod et al.: Microlensing variability in the Einstein Cross 5 Fig. 7.
Deconvolved and extracted 1D-spectra of the lensinggalaxy. The two panels correspond to the two MOS masks.The shaded areas are the envelopes containing all the spectraof the lens obtained with the corresponding mask. The thickblack lines are the means. Note the small scatter between thetwo spectra.We created a ratio spectrum for each star, i.e. we divided thespectrum of the star by the spectrum of the same star for a cho-sen reference exposure. We chose epoch − − ff erent stars arecompatible (see Fig. 4). If not, we rejected one or a maximumof two of the PSF stars. This can happen in some exceptionalcases, e.g. when the alignment between the star and the slit isnot optimal and generates a color gradient in the spectrum ofthe misaligned object. Aside from this instrumental e ff ect, theobservations show no trace of intrinsic variability of the PSFstars.After checking that the correction spectra obtained for thefour stars were very similar, we computed their mean, whichwe took as the correction to be applied to the gravitational lens.The high stability of the corrections across the field demon-strates that all residual chromatic slit losses due to the atmo-spheric refraction are fully corrected. This correction is easedby the fact that: (1) the position angle of the masks is the samefor the quasar images and for the PSF stars (i.e. the PSF clip-ping is the same for the target and the reference stars); (2)we avoid observations at large airmasses (i.e. never larger than2.5); and (3) the atmospheric refraction corrector on FORS1 isvery e ffi cient. Fig. 8.
Example of a spectral decomposition. The top panelshows the two extracted spectra for the images A and D forthe observations taken on epoch − − The lensing galaxy in QSO 2237 + / arcsec in the R band, which is comparable to thequasar images. Hence studying microlensing variations of thequasar images requires very accurate deblending.In order to carry out this challenging task, we used the spec-tral version of the MCS deconvolution algorithm (Magain et al.1998, Courbin et al. 2000), which uses the spatial informationcontained in the spectra of the PSF stars. The algorithm sharp-ens the spectra in the spatial direction, and also decomposesthem into a “point-source channel” containing the spectra of thetwo quasar images, and an “extended channel” containing thespectrum of everything in the image that is not a point source,in this case, the spectrum of the lensing galaxy. In Fig. 5, we il-lustrate the process and the di ff erent outputs. In Fig. 7, we showhow similar the spectra of the lensing galaxy are, extracted ei-ther from two di ff erent masks or from data taken at di ff erentepochs, hence illustrating the robustness of the deconvolutiontechnique. In Fig. 8, we give an example of decomposition ofthe data into the quasar and lens spectra after integrating alongthe spatial direction. The lensing galaxy spectrum shows notrace of the residual quasar BELs. Even when the contrast be-tween the quasar and the galaxy is particularly large, the de-composition is accurate. For example, the CaII H + K doublet in
A. Eigenbrod et al.: Microlensing variability in the Einstein Cross
Fig. 5.
Left: portion of the VLT 2D-spectrum of quasar images D and A, taken on epoch − − Center left: spatially deconvolved spectrum. The two quasarimages are very well separated.
Center right: spectrum of the lensing galaxy alone.
Right: residual map of the deconvolutionafter subtraction of the quasar and lens spectra. Note that the residuals are displayed with much narrower cuts than those usedin the other panels. The darkest and brightest pixels correspond to − σ and + σ respectively. No significant residuals of thespectral features are visible. Fig. 6.
Deconvolved and extracted spectra of quasar image A for five observing epochs. Chromatic variations in the spectra areconspicuous with the blue part of the spectra being more magnified than the red part.the lens spectrum is well visible, in spite of the presence of thestrong quasar C IV emission in the same wavelength range.
After reduction and spatial deconvolution, we obtained theextracted spectra of quasar images A and D on 30 di ff erent . Eigenbrod et al.: Microlensing variability in the Einstein Cross 7 Fig. 9.
OGLE-III light curves (Udalski et al. 2006) of all fourquasar images from April 2004 to December 2006 (dots),compared with the photometry derived by integrating ourVLT spectra through the OGLE V-band (dark triangles). The1 − sigma error bars correspond to the photon noise in thespectrum. We shift the OGLE-III light curve of image D by − . ff erent epochs. Several ex-tracted spectra of image A are shown in Fig.6. As a sanitycheck, we compared our results with the OGLE-III photomet-ric monitoring of QSO 2237 + V -band toestimate, from the spectra, the photometric light curves as ifthey were obtained from imaging. In Fig. 9, we compare ourmagnitude estimates with the actual OGLE-III measurements.The overall agreement is very good for images A, B, and C.For image D, we have to shift the OGLE-III light curve by − . − .
4. Multi-component decomposition Di ff erent emission features are known to be produced in re-gions of di ff erent characteristic sizes. As microlensing magni-fication varies on short spatial scales, sources of di ff erent sizesare magnified by di ff ering amounts (e.g. Wambsganss et al.1990). Emission features from smaller regions of the source aremore highly variable due to microlensing than features emittedin more extended regions. In order to study the variation ofeach spectral feature independently, we need to decompose thespectra into their individual components. In our analysis of the 1-D spectra of the four quasar images, wefollow the multi-component decomposition (MCD) approach(Wills et al. 1985, Dietrich et al. 2003) implemented in Sluseet al. (2007). This method is applied to the rest-frame spectra,assuming they are the superposition of (1) a power law con-tinuum, (2) a pseudo-continuum due to the merging of Fe IIand Fe III emission blends, and (3) an emission spectrum dueto the other individual BELs. We consider the following emis-sion lines : C IV λ λ λ λ λ λ λ λ ≤ λ ≤ ≤ λ ≤ f ν ∝ ν α ν , which translates in wave-length to f λ ∝ λ α λ with the relation α ν = − (2 + α λ ), i.e. f λ = f λλ ! α λ = f λλ ! − (2 + α ν ) where λ = α ν is the commonly used canon-ical power index.Next, we fit the BELs with Gaussian profiles. We considera sum of three profiles to fit the absorption feature in the C IVemission line. Two profiles are used for the C III] line and onesingle profile is used to fit simultaneously the O III] and Al IIlines. All other BELs are fitted with one single profile. Wethen subtract the BELs and the continuum from the spectra.We consider the residuals as coming from the emission blendsof Fe II and Fe III. Hence the averaged and normalized residu-als over all epochs define our first iron pseudo-continuum tem- A. Eigenbrod et al.: Microlensing variability in the Einstein Cross
Fig. 10.
Multi-component decomposition of the spectrum of the brightest quasar image, A, taken on epoch − − σ ). Fig. 11.
Left:
Examples of light curves for the quasar image A. The integrated flux for the continuum, the He II, and C III] BELsare given from top to bottom. The continuum is integrated over the entire available wavelength range. In each panel, we fit a scaledversion (solid line) of the OGLE-III light curve (Udalski et al. 2006). This nicely illustrates that the BELs vary simultaneouslyand proportionally to the continuum.
Right: variability of the best-fit parameters α ν and f of the continuum (see Section 4.1).Note how the changes in slope of the continuum ( α ν ) follow those of the mean intensity ( f ). . Eigenbrod et al.: Microlensing variability in the Einstein Cross 9 Fig. 12.
Correlation between the intensity f and slope α ν ofthe continuum spectra for all four quasar images. The pointsare connected chronologically. The first observation epoch ismarked by a star and the last one by a square. The correla-tion is obvious in images A and B spanning a broad range ofspectral slopes. In these two images, an increase in intensityis accompanied by an increase in steepness, i.e. when a quasarimage gets brighter, it also gets bluer.plate. We can then proceed iteratively by including this pseudo-continuum iron template in the fitting procedure and rerun it.This gives a better fitting of the emission lines and defines animproved iron pseudo-continuum template. After five of theseiterations, typically the fitting does not change significantlyanymore. Fig. 10 shows an example of the fitting decompo-sition described above. The light curves for the continuum and for the emission linescan be constructed from the above multi-component decom-position. We show in Fig. 11 an example of variation in thebrightest quasar image, A, for the continuum and for two BELs.The error bars give the photon noise, integrated over the corre-sponding wavelength range. In the right panel of Fig. 11, weshow the variability of the continuum in intensity, f , and inslope, α ν , for the 4 quasar images. It is immediately clear thatthe continuum variations with the largest amplitude are ob-served in image A, between HJD = = f and α ν are strongly correlated for images A and B, indicating that significant microlensing eventsoccured in these two images.Inaccurate alignment of the quasar images in the slit of thespectrograph is a possible instrumental e ff ect that can mimicmicrolensing changes in the spectral slope of the quasar im-ages. Indeed, small clipping of one of the quasar images wouldlead to a stronger flux loss at the bluer wavelengths, henceproducing a color gradient in the spectrum and a decrease inthe measured value α ν . We have checked all the “through-slit”images taken before each spectrum. Not only do these imagesshow that the alignment is correct, but it is also very easily re-producible from one epoch to another, even when the FORS1has been dismounted from and remounted on the telescope.We have also checked that our fitting procedure does not in-troduce any spurious correlation between α ν and f . We checkthis by using simulated spectra. In order to do that we take areference spectrum for each quasar image and subtract its con-tinuum. We then take random pairs of ( α ν , f ) parameters sothat the α ν vs. f plane is well sampled. We chose 400 suchpairs and create the corresponding continuum to be added tothe reference spectrum. The decomposition procedure is thenrun on the 400 spectra. We find no correlation at all betweenthe measured α ν and f . In addition, the parameters used tobuild the simulated spectra are almost perfectly recovered bythe decomposition procedure.We conclude that genuine chromatic variations are presentin the continuum of all images of QSO 2237 + ff ectis most pronounced in image A during the last observing sea-son, and in image B at the beginning of our monitoring. Weshow in the following that these observed variations are, in ad-dition, well compatible with the OGLE-III single-band photo-metric observations.
5. Microlensing variability in the OGLE-IIIphotometry
The photometric variations in most gravitationally-lensedquasars are dominated by the intrinsic variations of the quasar,typically of the order of 0 . − . . − . +
434 by Burud et al.2000; RX J0911 + + + ff erent from this general behaviorin two ways: (1) the time delays between each pair of imagesare expected to be of the order of one day, hardly measurable,and (2) the microlensing variations dominate the light curves.For these two reasons, microlensing can be fairly well isolatedin each quasar image, because it acts di ff erently on the foursightlines.To separate the intrinsic flux variations of the quasar fromthe microlensing ones, we perform a polynomial fit to theOGLE-III light curves (Udalski et al. 2006) of Fig. 9. Thissimple and fully analytical method has been developed byKochanek et al. (2006), and is also described by Vuissoz et al.(2007). In the present application, the variations of each quasar Fig. 13.
Decomposition of the OGLE-III photometric lightcurves (Udalski et al. 2006) of the quasar images, into intrin-sic quasar variations and microlensing-induced variations (seeSection 5). The intrinsic variations are shown at the bottom ofthe figure as a continuous line, while the pure microlensingvariations are the data points. The curves are shifted arbitrar-ily along the y-axis for clarity. The tickmarks at the top showthe epochs of our observations.image are modeled as a sum of two Legendre polynomials: onepolynomial is common to all four quasar images and representsthe intrinsic variations of the source while a second polyno-mial, di ff erent for each quasar image, represents the additionalmicrolensing variations. In doing so, we rescale the OGLE-IIIerror bars of each image by a factor equal to the flux ratio be-tween each image and image C. This rescaling suppresses thepotential problem existing if the fitting procedure considers thevariation of image A (with the highest signal-to-noise) as theintrinsic variation of the quasar. The chosen order of the poly-nomial is 7 for the intrinsic variation, and 10 for the microlens-ing variation. Higher order polynomials do not significantly im-prove the fit. The results are displayed in Fig. 13, where the in-trinsic variation of the source recovered by the simultaneous fitis shown together with the pure microlensing variations.We check the e ffi ciency of our method by generating arti-ficial light curves and then using the above polynomial fit torecover the intrinsic and microlensing light curves. These arti-ficial light curves are generated in the same way as describedin Eigenbrod et al. (2005), and are composed of an intrinsiclight curve to which we add microlensing fluctuations. Bothare created in a random walk manner (i.e. not from polynomi-als). They are constructed to match the variability propertiesof the actual light curves, i.e. their timescale and amplitude ofvariation (for further details see Eigenbrod et al. 2005). We re- cover the simulated intrinsic light curve of the quasar with atypical error of less than 0.1 mag. The variations of more than0.4 mag, shown in Fig. 13, both for microlensing and quasarvariations, are well above the error estimated from the simu-lated light curves. In our simulations, we adopt the same pho-tometric error bars as in the light curves of all quasar images,i.e. the re-scaling of the error bars described above in the realdata is taken into account. If, on the contrary, we adopt errorbars that follow the photon noise, the fitting procedure consid-ers the highest signal-to-noise light curve as the intrinsic quasarlight curve.The light curve most a ff ected by microlensing is that ofimage B, with a peak-to-peak amplitude of more than 0.7 magover 3 years. The other quasar images show microlensing-induced variations of up to 0.4 mag, with quasar image A hav-ing a sharp event during the last observing season. The intrinsicquasar light curve displays a variation of about 0.4 mag.The polynomial decomposition of the light curves are com-patible with the spectroscopic results. Quasar images A and B,which have the largest microlensing contribution in Fig. 13, re-spectively at HJD ∼ ∼ α ν at the same epochs.
6. Microlensing variability in the spectra
Chromatic variations of the continuum of images A and B ofQSO 2237 + ff ects havealready been observed by Lewis et al. (1998) and Wayth et al.(2005), but only for data over two epochs. Our VLT spectra al-low us to follow the variations over two full years, provided theintrinsic variations of the quasar are removed. Since the time delays in QSO 2237 + F ( t ) be the intrinsic sourceflux, and M i , µ i be the macro and microlensing-magnificationsof quasar image i , respectively. The observed flux ratio betweenimages i and j at time t is then : R i j ( t ) = µ i ( t ) M i F ( t ) µ j ( t ) M j F ( t ) e − ( τ i − τ j ) = µ i ( t ) M i µ j ( t ) M j e − ( τ i − τ j ) . (1)The extinction e − τ i remains constant in time and is rela-tively similar in all four quasar images as we show in Sec. 6.3.Hence, it is not expected to strongly a ff ect our results and wewill neglect it in the following. The macromagnifications M i are best estimated in the mid-IR and radio domain (Falco et al.1996, Agol et al. 2000). At these wavelengths, the source sizeis much larger than the typical spatial scale in the microcaus-tics network, hence leaving it fairly una ff ected by microlensing(i.e. µ i = R i j by M j / M i , using the mid-IRobservations, we find the pure-microlensing magnification ra-tios: r i j ( t ) = µ i ( t ) µ j ( t ) . (2) . Eigenbrod et al.: Microlensing variability in the Einstein Cross 11 Fig. 14.
Microlensing-magnification ratios r i j as a function of time (HJD − In Fig. 14, we show the variations of r i j ( t ) for the integratedflux in the main emission lines and in the continuum. We plotin the figure all possible combinations of image ratios. Severalinteresting results can be drawn from this figure:(a) With the exception of r BD during the last season (i.e.HJD > r i j are close to1, meaning that microlensing is acting at least in threequasar images during the entire monitoring. The evidencefor r BD ∼ > r i j , r BD ). The r i j ratios inthe BELs generally closely follow the value measured forthe continuum. This demonstrates that the BLR is smallenough - probably not larger than a few Einstein radii ofa typical microlens in QSO 2237 + ff ected. In addition, the variations ob-served in the BELs are correlated with those in the contin-uum.(c) The largest changes of magnification ratios involve imagesA (for HJD ∼ ∼ r A j and r C j deviate significantly from 1 duringthe whole observing period. As an example, we measurein the continuum (resp. C III]) r AD ∼ . < r CD ∼ . ∼ r AD and r CD are also similar to the value mea-sured in the H β emission line in August 2002 (HJD = ff ected by long-term micro / milli-lensing on periods longer than 5 years.(e) The ratio r CD is the most stable ratio along the monitor-ing campaign, indicating that no major (short) microlensingevent occured in images C or D.Because of (b) and (e), we can safely consider that duringthe time span of our observations, image D is the less a ff ectedby microlensing, both on short (i.e. of the order of a few weeks)and long timescales (years). This is consistent with the broad-band microlensing light curves of Fig. 13.We use image D as a reference to study the “short” mi-crolensing events a ff ecting image A at HJD ∼ ∼ r AD and r BD during microlensing events and in more “quies-cent” phases. For image A (i.e. r AD ), we clearly see in this tablethat C IV, C III] and He II show very similar magnification ra-tios, while the Mg II line is less magnified.All the lines are less magnified than the continuum, consis-tent with a scheme where the continuum is emitted in the mostcompact region, and where other emission lines are emitted inlarger regions, the largest region being the one with the lowest Table 2.
Mean microlensing ratios for the continuum and forthe main BELs. The mean values for r AD = µ A /µ D and r BD = µ B /µ D are computed for the observations around the epoch inthe HJD line, i.e. during microlensing events or during quies-cent phases. The values are given along with the dispersion ofthe points around the mean. < r AD > < r AD > < r BD > < r BD > HJD 3900 d 3500 d 3500 d 3300 dState Micro-A Quiet-A Micro-B Quiet-BCont. 3.46 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± + III 2.56 ± ± ± ± ionization potential (i.e. Mg II). Indeed the ionization poten-tials of the di ff erent lines are 47.9 eV (C IV), 24.6 eV (He II),24.4 eV (C III]), and 7.6 eV (Mg II). The di ff erence of magni-fication between C III] and Mg II was not observed by Waythet al. (2005). For image B, the same global trend is observedexcept that the relative errors on the r BD ratios are higher dueto the lower signal-to-noise ratio of the spectra of image B. Thee ff ect is almost absent when r AD and r BD are computed in qui-escent phases of components A and B.The behaviour of the Fe II + III emission is more di ffi cultto interpret, as this complex is in fact a blend of many lines.However, we note that the Fe II + III complex in image A is mi-crolensed at about the same level as the Mg II line. In addition,the di ff erence in magnification of the Fe II + III lines betweena microlensing and a quiet phase is larger than for the otherlines. This may suggest di ff erential magnification of the emit-ting regions within the Fe II + III complex, i.e. that the Fe II + IIIis present both in compact and more extended regions, a con-clusion also reached by Sluse et al. (2007).
We have discussed in the previous sections the global inten-sity changes in the emission lines of QSO 2237 + ff erential magni-fication of regions with di ff erent velocities in the BLR. Thiscan introduce, e.g., asymmetric changes or even peak dis-placement of the line (Lewis 1998, Abajas et al.2002, Lewis& Ibata 2004). Observational evidence for profile variationshas been reported, e.g. in HE 2149-27 (Burud et al. 2002a);SDSS 1004 + + − . Eigenbrod et al.: Microlensing variability in the Einstein Cross 13 Fig. 15.
The C IV (left) and C III] (right) BEL profiles for quasar images A, B, C and D as observed at epoch − − − − ∼ − − − and 5 000 km s − . We will refer to thesecomponents in the following as the “narrow” and “broad” com-ponents, although the di ff erence made here between narrowand broad is only phenomenological. In particular, the narrowcomponent is not associated, a priori , with the narrow line re-gion (NLR).We measure the microlensing magnification ratio µ i (narrow) /µ i (broad) for the C III] emission line in eachquasar image. In image A, the magnification of the broadcomponent is ∼ β emission lines(Metcalf et al. 2004). This gives a hint that the narrow compo-nent of the C III] emission line may be partly emitted by theNLR.However, we do not measure the same microlensing-magnification ratios r i j in the narrow C III] emission as thosemeasured by Metcalf et al. (2004) in the narrow [O III] emis- sion line. The r BC , r BD , r CD ratios measured by these authorsare similar for the H β BEL and for the narrow [O III] emis-sion line, but are di ff erent from 1, which may be interpretedas a consequence of the microlensing of the NLR. If the NLRis indeed microlensed, the discrepancy between our values andthe ones by Metcalf et al. (2004) can simply be explained bythe fact that the amplitude of microlensing changes with time.However, there are also two other possible explanations. First,the size of the [O III] emission region might be comparablewith the size of the macrocaustic of the lensing galaxy, suchthat the [O III] lensed images are extended and can be resolvedby integral field spectroscopy (Metcalf et al. 2004, Yonehara2006). This may lead to uncertainties in the flux measurementsdepending on the chosen size of the aperture. Finally, the dis-crepancy may be explained as well in terms of extinction by thelens (see Sec. 6.3).The details of the C III] profile also show variations. Ourspectra show evidence for systematic broadening, by ∼
10% ofthe C III] emission line in quasar image A during our last ob-serving season (HJD > Fig. 16.
Comparison of the C III] BEL profiles of images A, B, C and D of the lensed quasar QSO 2237 + ff erentepochs. The line profiles are normalized so that they have the same peak value. This representation illustrates the di ff erence inthe line profiles of the 4 lensed images at several epochs. The most striking e ff ect is the larger wings observed at all epochs inimage A. These larger wings are caused by di ff erential microlensing between the wings and the core of the C III] line. Smallmicrolensing induced fluctuations of the lines profiles are also observed in images B, C, and D. The epochs are chosen to samplethe whole observing period. . Eigenbrod et al.: Microlensing variability in the Einstein Cross 15 lines at a single epoch would match perfectly. It is conspicuousin Fig.16 that this is not the case. First, we clearly see the e ff ectof di ff erential magnification between the core and the wingsof image A at all epochs (the wings of C III] in image A arealways larger than in the other emission lines). The emissionlines in B, C, and D are more similar to one another but there isno epoch where the three line profiles match perfectly, indica-tive that small microlensing fluctuations a ff ect the C III] line.Image C also shows line profile variations, even though the ef-fect is less pronounced than in image A. Finally, we note theabsence of any strong line profile variations of image B dur-ing the short-term microlensing event occuring in that image atHJD ∼ Most of the magnification ratios r i j fluctuate around a meanvalue. The fluctuations themselves can only be explained bymicrolensing, but the value of the mean < r i j > is usually dif-ferent from 1 during our observations. This can, in principle, beexplained either by long-term microlensing or by non variableextinction by dust in the lens.In order to test the latter hypothesis, we use a simple andempirical diagnostic using ratio spectra of pairs of quasar im-ages. In the absence of reddening and microlensing, these ratiospectra should be flat. If dust is present in di ff erent amounts onthe lines of sight of the images, the ratio spectra will show anon-zero slope, constant with time. Any time-variable changeof slope can safely be attributed to microlensing.We find that the most useful pairs are the ones formed byA&C, B&C, and C&D. Indeed, the C / D ratio spectrum is foundto be almost flat all along the years, indicating no significantdi ff erential extinction. This is not the case for the two otherpairs of ratio spectra which show reddening of image C relativeto both A and B, as also reported by Yee (1988). We estimatethis di ff erential reddening using the extinction law by Cardelliet al. (1989) and by assuming R V = .
1. We find that the dif-ferential extinction A V ( C ) − A V ( A ) ≃ A V ( C ) − A V ( B ) is in therange 0.1-0.3 mag. This range of values is su ffi cient to explainthe discrepancy found between the r i j ratios measured in theC III] narrow component and in [O III] of Sec. 6.2.Finally, our estimates of the extinction in the lens are toosmall to explain the highest values of the mean magnificationratios observed in Fig. 14 and Table 2. For instance, the meanof the r AB or r AD ratio reaches values larger than 2. Static, long-term microlensing is therefore present in the Einstein Cross, atleast in images A and C.
7. Conclusions
This paper presents the first long-term (2.2 years) and well-sampled spectrophotometric monitoring of a gravitationallylensed quasar, namely the Einstein Cross QSO 2237 + + ff ectedby long-term microlensing on a period longer than 5 years.This long-term microlensing a ff ects both the continuum andthe BELs.Furthermore, in quasar image A, the broad component ofthe C III] line is magnified by a factor 1.8 larger than the narrowcomponent. On the contrary, the other quasar images have thesame magnification in the narrow and broad components.On the short timescales, i.e. several months, images A andB are the most a ff ected by microlensing during our monitor-ing campaign. Image C and especially D are the most qui-escent. Image A shows an important brightening episode atHJD ∼ ∼ ∼ ff erent sizes are responsiblefor the Fe II + III emission.Finally, we estimate the di ff erential extinction betweenpairs of quasar images due to dust in the lensing galaxy to bein the range 0.1-0.3 mag, with images C and D being the mostreddened. This amount of di ff erential extinction is too small toexplain the large microlensing-magnification ratios involvingimages A and C. Long-term microlensing, beyond the durationof our observations, is therefore present in these images.The timescales of the microlensing variations inQSO 2237 + ffi cient tosample the events well enough. In addition, the Einstein Crossis the lensed quasar with the fastest and sharpest microlensingevents. It is therefore unique in the sense that only a fewyears of monitoring can truly constrain the quasar structure onparsec scales (Kochanek 2004). In addition, in the case of the Einstein Cross, the very di ff erent behaviours of the BELs andthe continuum with respect to microlensing o ff er considerablehope to reconstruct the two types of regions independently,using ray-shooting simulations.With two more years of data, we expect to map a total of upto half a dozen microlensing events in the four quasar images,hence providing a unique and useful data set for microlensingand quasar studies. Acknowledgements.
We are extremely grateful to all ESO sta ff fortheir excellent work. The observations presented in this article haveinvolved a lot of e ff orts from the ESO sta ff operating FORS1, toensure accurate and reproducible mask alignment, to keep the bestpossible temporal sampling, and to meet the requested seeing value.This project is partially supported by the Swiss National ScienceFoundation (SNSF). References
Abajas, C., Mediavilla, E., Mu˜noz, J. A., et al. 2002, ApJ, 576, 640Agol, E., & Krolik, J. 1999, ApJ, 524, 49Agol, E., Jones, B., & Blaes, O. 2000, ApJ, 545, 657Burud, I., Hjorth, J., Jaunsen, A. O., et al. 2000, ApJ, 544, 117Burud, I., Courbin, F., Magain, P., et al. 2002, A&A, 383, 71Burud, I., Hjorth, J., Courbin, F., et al. 2002, A&A, 391, 481Cardelli, J. A., Clayton, G. C., Mathis, J. S. 1989, ApJ, 345, 245Chartas, G., Eracleous, M., Agol, E., et al. 2004, ApJ, 606, 78Colley, W. N. & Schild, R. E. 2003, ApJ, 594, 97Courbin, F., Magain, P., Kirkove, M., & Sohy, S. 2000, ApJ, 529, 1136Dietrich, M., Hamann, F., Appenzeller, I., et al. 2003, ApJ, 596, 817Eigenbrod, A., Courbin, F., Vuissoz, C., et al. 2005, A&A, 436, 25Eigenbrod, A., Courbin, F., Dye, S., et al. 2006a, A&A, 451, 747Eigenbrod, A., Courbin, F., Meylan, G., et al. 2006b, A&A, 451, 759Falco, E. E., Lehar, J., Perley, R. A., et al. 1996, AJ, 112, 897Filippenko, A. V. 1989, ApJ, 338, L49G´omez- ´Alvarez, P., Mediavilla, E., Mu˜noz, J. A., et al. 2006, ApJ,645, 5G´omez- ´Alvarez, P., Mediavilla, E., S´anchez, S. F., et al. 2004, AN,325, 132Hjorth, J., Burud, I., Jaunsen, A. O., et al. 2002, ApJ, 572, 11Huchra, J., Gorenstein, M., Kent, S., et al. 1985, AJ, 90, 691Irwin, M. J., Webster, R. L., Hewett, P. C., et al. 1989, AJ, 98, 1989Jakobsson, P., Hjorth, J., Burud, I., et al. 2005, A&A, 431, 103Kaspi, S., Smith, P. S., Netzer, H., et al. 2000, ApJ, 533, 631Keeton, C. R., Burles, S., Schechter, P. L., et al. 2006, ApJ, 639, 1Kent, S. M., & Falco, E. E. 1988, AJ, 96, 1570Kochanek, C. S. 2004, ApJ, 605, 58Kochanek, C. S., Morgan, N. D., Falco, E. E., et al. 2006, ApJ, 640,47Koptelova, E., Shimanovskaya, E., Artamonov, B., et al. 2005,MNRAS, 356, 323Lewis, G. F., 1998, MNRAS, 297, 69Lewis, G. F., Irwin, M. J., Hewett, P. C., et al. 1998, MNRAS, 295,573Lewis, G. F. & Ibata, R. A. 2004, MNRAS, 348, 24Magain, P., Courbin, F., & Sohy, S. 1998, ApJ, 494, 452Metcalf, R. B., Moustakas, L. A., Bunker, A. J., et al. 2004, ApJ, 607,43Mineshige, S., & Yonehara, A. 1999, PASJ, 51, 497Murray, N., Chiang, J., Grossman, S. A. & Voit, G. M. 1995, ApJ, 451,498Nemiro ffff