FERMI LARGE AREA TELESCOPE DETECTION OF GRAVITATIONAL LENS DELAYED γ-RAY FLARES FROM BLAZAR B0218+357

B0218+357 was discovered with the NRAO 140 ft telescope in its strong source survey (S3 0218+35; Pauliny-Toth & Kellermann 1972). Later radio imaging revealed it to be a gravitationally lensed blazar with the smallest separation double-image known (335 mas) and an Einstein ring with a similar angular diameter (O'Dea et al. 1992; Patnaik et al. 1993). The lens galaxy is at redshift z = 0.6847 (Browne et al. 1993), and the blazar was later securely measured at z = 0.944  ±  0.002 (Cohen et al. 2003).

Shortly after the lens discovery, Corbett et al. (1996) measured a time delay (Refsdal 1964) Δt_r = 12 ± 3 days (1σ quoted throughout unless otherwise specified) at radio wavelengths, using the Very Large Array (VLA) to spatially separate and monitor the polarization variability in its leading brighter A (western) and fainter B (eastern) images. Later independent (but contemporaneous) dual-frequency VLA observations further refined the delay, Δt_r = 10.5 ± 0.2 (Biggs et al. 1999) and 10.1 ± 0.8 days (Cohen et al. 2000). Interestingly, Eulaers & Magain (2011) analyzed the latter's measurements and found two possible delays, ${\Delta t_{\rm r}}=9.9^{+4.0}_{-0.9}$ or 11.8 ± 2.3 days. Although these delays span a narrow range, Δt_r ∼ 10–12 days, because of the differing assumptions and analysis techniques employed in these works, there remains some debate regarding how to best derive their uncertainties.

B0218+357 is also a γ-ray source detected by the Fermi Large Area Telescope (LAT; Atwood et al. 2009) with an average flux²⁸ F_γ = (1.00 ± 0.07) × 10⁻⁷ photons cm⁻² s⁻¹ over its first two years of observations (2FGL J0221.0+3555; Nolan et al. 2012). Its steep spectrum at >100 MeV energies (photon index, Γ = 2.28 ± 0.04) and overall spectral energy distribution are typical of an otherwise normal γ-ray emitting flat-spectrum radio quasar (e.g., Ghisellini et al. 2010). While γ-ray data lack the necessary spatial resolution to separate lensed images, such blazars display their most dramatic variability in γ-rays, and the LAT's all-sky monitoring could give it a distinct advantage over lower-frequency imaging observations in parameterizing lensed systems. Indeed, Atwood (2007) proposed prior to Fermi's launch that the LAT could detect delayed emission from such gravitationally lensed blazars using integrated light curves for sufficiently bright γ-ray flares. B0218+357 was found to be variable in the early LAT observations, though only modestly so (Abdo et al. 2010; see Figure 1).

Zoom In Zoom Out Reset image size

Bright γ-ray flaring from B0218+357 was detected with the LAT beginning late 2012 August (Ciprini 2012), and a delayed flare was tentatively identified ∼10 days later (Giroletti et al. 2012), consistent with the radio delay measurements. The blazar then displayed even brighter, more sustained flaring activity beginning September 14, thus prompting a Fermi target of opportunity (ToO) pointed observation (Cheung et al. 2012) that traced the anticipated delayed emission in detail. Two main additional flaring events were subsequently observed in as many months (see Figure 1 for an overview). We discuss the temporal and spectral γ-ray properties of B0218+357 together with the derived time lag, flare timescales, and observed flux ratios of the A/B images.

The Fermi-LAT operates in default sky-survey mode, and over every two ∼1.6 hr spacecraft orbits, provides observations covering the entire sky. We used LAT observations with the P7SOURCE_V6 instrument response functions, selecting 100 MeV–100 GeV events with a region of interest (ROI) of radius = 15° centered at the B0218+357 radio position, R.A. = 3527279, decl. = 3593715 (J2000; Patnaik et al. 1992). The maximum zenith angle of 100° was set to minimize the contamination from Earth limb photons as well as the appropriate gtmktime filter (No. 3) following the FSSC recommendations²⁹ for the combination of sky-survey and pointed observations. The gtlike likelihood in the Fermi Science tools (version v9r27p1) was used for the spectral analysis, assuming throughout a single power-law model for B0218+357 over the selected energy range (as in the 2FGL catalog). The background model included all 2FGL sources within the ROI as well as the Galactic (gal_2yearp7v6_v0.fits) and isotropic (iso_p7v6source.txt) diffuse components.

In generating each light curve, the isotropic normalization was left free to vary in each time bin while the two known variable 2FGL sources within a 5° ROI and the Galactic normalization were initially fitted over each full interval, then fixed at the average fitted values in the shorter time bins. As a convenient reference point, we define T = MJD − 56100 days (i.e., T = 0 was 2012 June 22), the time when γ-ray flaring became obvious. Integrating 1417 days (∼3.9 yr) of LAT observations prior to this date gave an average F_γ = (0.83 ± 0.05) × 10⁻⁷ photons cm⁻² s⁻¹, with Γ = 2.30 ± 0.03, consistent with the 2FGL value. For context, we generated a one week binned light curve for five years of data (2008 August 5–2013 August 6; Figure 1, top) assuming a fixed Γ = 2.3. Besides the modest source activity in early 2009 and 2010, the pronounced flaring beginning in mid-2012 lasting for ∼200 days is apparent; thereafter, the source quieted again to earlier levels.

In order to study the flaring activity in detail, we defined a 265 day interval starting at T = 0 days and generated 1 day and 6 hr binned light curves. The Fermi ToO observations also allowed us to produce a ∼1.6-hr orbit-by-orbit binned light curve for the sub-interval covering the first delayed flare from 2012 September 24–October 1 (T = 94–101 days). To search for any possible spectral changes, we initially computed the 1 day binned light curve with the photon index free in the fit. For the 108 points with the greatest significance (test statistic,³⁰TS ⩾ 25), we found all but four points within 2σ of the weighted average value of 2.31 ± 0.02, which in turn is consistent with the 3.9 yr average. We thus regenerated the 1 day (Figure 1, middle), the 1.6 hr orbit (Figure 1, bottom), and the 6 hr binned light curves (Figure 2) with Γ = 2.3 fixed.

Zoom In Zoom Out Reset image size

3.1. Time Lag

The B0218+357 γ-ray light curve appears quite complex with many peaks and valleys over the ∼4 months from T ∼ 60–180 days (Figure 2) when the source was most active. To search for a time lag, we computed the auto-correlation function (ACF) for the 6 hr binned light curve up to lag values of half of the total defined 265 day flaring interval. This evenly sampled light curve consisted of 1057 measurements with three missing data points due to exposure gaps. The ACF was therefore computed both by a standard (after interpolating the three missing points) and a discrete routine (Edelson & Krolik 1988). The two procedures gave almost identical results and the ACF is shown in Figure 3. A single prominent correlation peak is apparent between the time lag range of 11–12 days. The peak's significance is 9σ with respect to the measurement noise and comparing it to the height above the ACF "background." Fitting a Gaussian function to this peak, we estimated a best-fit value, Δt_γ = 11.46  ±  0.16 days (1σ). Uncertainties were estimated by a model independent Monte Carlo method (Peterson et al. 1998) accounting for the effects of measurement noise and data sampling. The time lag does not match any known period observed with the LAT (Ackermann et al. 2012; Corbet et al. 2012). Because the γ-ray flaring was so pronounced especially from T ∼ 84–155 days, and appears to be broadly divided into three ∼8–10 day-long flare/delay sequences (Section 3.2), this could induce other smaller enhancements in the ACF over the studied interval.

Zoom In Zoom Out Reset image size

As a cross-check of the lag derived from the full flaring interval γ-ray data, discrete ACFs were computed for two segments from T = 0–110 and T = 110–265 days. The lags obtained from Gaussian fits to the peaks were Δt_γ = 11.52 ± 0.31 and Δt_γ = 11.38 ± 0.28 days, respectively, confirming the delay value and small uncertainty for the full interval, thus indicating that we obtained a robust measurement with the LAT. The small uncertainty in Δt_γ is comparable to the best determined radio measurements for B0218+357 although the former is marginally larger by Δt_γ − Δt_r = 1.0 ±  0.3 and 1.4  ±  0.8 days (1σ) than the Biggs et al. (1999) and Cohen et al. (2000) values, respectively, but consistent with the Eulaers & Magain (2011) values. If the radio/γ-ray delays are intrinsically different due to an offset between the respective emitting regions, the implied offset in a singular isothermal sphere lens model is ∼70 pc (projected) for a ∼10% difference in the time delay. This seems extreme considering such offsets are on average ∼7 pc in other blazar jets (e.g., Pushkarev et al. 2010), and may rather suggest the uncertainty in the radio delay was underestimated (Section 1).

3.2. Flare Timescales

3.3. Flux and Magnification Ratios

Adopting the γ-ray delay, we compared the 6 hr binned light curves for the three main flaring episodes with the observations shifted by −11.46 days and computed the observed ratios between corresponding flux pairs, retaining only ratio values ⩾2 × their uncertainties (Figure 4). The first sequence appears to show the clearest correspondence between features in the two light curves, with only minor deviations about the weighted average flux ratio 1.3 ± 0.1. By subtracting a baseline, F_γ = 0.3 × 10⁻⁶ photons cm⁻² s⁻¹ (the minimum observed flux during the overall flaring interval), we can further estimate a corresponding magnification ratio in γ-rays of ≈1.3, consistent with the flux ratio. The average ratios for this first sequence seem to imply the brighter A image led the B image in γ-rays, as observed in the radio. More conservatively, however, given the large uncertainties in the individual measurements, the flux ratios appear consistent with unity. Moreover, for the subsequent second and third sequences, the correspondences between the flare and delayed emissions were less clear. Sharp and more scattered changes in the paired flux ratios were apparent, including values <1 (which would imply a fainter leading A image). We interpret this as an artifact due to contamination from superposing flares after the source has already entered a very active phase. This confusion in the integrated light curves prevents us from reliably determining magnification ratios, and how variable this quantity may have been.

Zoom In Zoom Out Reset image size

The flux ratio measured in γ-rays is smaller than in the radio. Biggs et al. (1999) found a small, but statistically significant frequency dependence in the flux ratios, 3.57 ± 0.01 (8 GHz) and 3.73 ± 0.01 (15 GHz), while Cohen et al. (2000) found similar values but with larger uncertainty, $3.2^{+0.3}_{-0.4}$ (8 GHz) and $4.3^{+0.5}_{-0.8}$ (15 GHz). Frequency dependence in the flux ratios of the two radio images and their observed substructures could be possibly due to free–free absorption and scattering from a molecular cloud in the lens galaxy (Mittal et al. 2007). We note that the radio and γ-ray observations are not simultaneous and magnification ratios could be variable with time. Further complicating such comparisons are open questions in blazar jet studies, i.e., the radio and γ-ray emitting regions need not coincide, with the latter likely more compact (Section 3.2), and whether successive γ-ray flares originated in a single emission zone or from separate relativistically moving dissipation regions along the jet. Excursions could also be due to intrinsic changes in the magnification ratios or microlensing from the relative motion of the source seen through a clumpy lensing galaxy. Indeed, microlensing in the context of extremely compact γ-ray emission zones (Torres et al. 2003) could explain the single 6 hr flare points that do not have corresponding lags (marked with asterisks in Figure 2), although fast superposed flares are also a possibility. Note that in optical and infrared observations, the B image appears brighter than the A image, i.e., reversed from the radio situation, and this is likely due to a combination of extinction of the A image and microlensing (Falco et al. 1999; Jackson et al. 2000).

Our detection of a gravitational lens time delay, Δt_γ = 11.46  ±  0.16 days, in the LAT observations of blazar B0218+357 has some interesting potential implications for future γ-ray studies. Foremost, the LAT detection of a γ-ray gravitational lens flaring event in B0218+357 suggests that such a measurement is possible in other blazars. In particular, gravitational lenses found in surveys of flat-spectrum radio sources (Browne et al. 2003; Winn et al. 2000) comprise a relevant sample as these form the basis of candidate γ-ray blazar catalogs (e.g., Healey et al. 2007). There are ∼20 gravitational lenses from these surveys out of >10⁴ radio sources studied with ≳ 30 mJy at 8 GHz and, so far, the two radio brightest are detected γ-ray sources PKS1830–211 (below) and B0218+357 (out of ∼10³ known γ-ray blazars; Nolan et al. 2012). The other fainter lensed systems are typically less variable at radio frequencies, making delay measurements difficult (e.g., Jackson 2007; Eulaers & Magain 2011) and while they are not yet reported γ-ray sources, the all-sky monitoring of Fermi-LAT will allow the detection of short-timescale flaring γ-ray activity in which to attempt delay measurements. Importantly, γ-ray measurements constrain lens parameters free of propagation effects like scintillation (Heeschen 1984; Lovell et al. 2008) that can hamper radio delay attempts (Winn et al. 2004), although microlensing may be an important limiting factor because γ-ray emitting regions are expected to be more compact than in the radio.

The case of B0218+357 appears to be the first clear case of a γ-ray detected gravitational lens time delay for any astrophysical system. Previously, γ-ray flaring from the gravitationally lensed z = 2.507 blazar PKS1830–211 was detected with the Fermi-LAT (Ciprini 2010) with a claimed delay, Δt_γ = 27.1 ± 0.6 days (Barnacka et al. 2011), consistent with the radio measurement, ${\Delta t_{\rm r}}=26^{+4}_{-5}$ days (Lovell et al. 1998). Subsequent analysis of more LAT data, including several prominent flares, did not confirm the γ-ray delay (Abdo et al. 2013). If the γ-ray delay in PKS1830–211 is assumed to be the same as the radio-measured delay, the non-detection of delayed γ-ray flares implies a magnification ratio in γ-rays much larger (≳ 6) than that observed in the radio (1.52 ± 0.05; Lovell et al. 1998), thus opposite of what we observed in B0218+357. With only two examples studied, no trend is clear. However, if microlensing effects can be disentangled (and in fact utilized as additional constraints on the emitting region source size), magnification ratios in radio and γ-ray arising from spatially distinct emission regions may be utilized as a probe of differing multi-frequency jet structures (see Martí-Vidal et al. 2013).

A time delay due to gravitational lensing of a background source by a foreground object can constrain Hubble's parameter (Refsdal 1964). The original lens model for B0218+357 (Biggs et al. 1999) predicted a delay, ${\Delta t}=7.2^{+1.3}_{-2.0}\,h^{-1}$ days (95% confidence). Utilizing improved localization of the lensing galaxy, the delay model uncertainty was reduced to 6.0% (York et al. 2005; see also Wucknitz et al. 2004), thus deriving h = 0.70 ± 0.05, assuming the often quoted Biggs et al. (1999) measured radio delay (see  Section 3.1). Adopting the York model for our independent γ-ray measured delay results in h = 0.64 ± 0.04, where this quoted uncertainty is due only to the time delay estimate and the statistical uncertainty in the mass model. Systematic errors in the modeling, and additional uncertainty due to line-of-sight structures (e.g., Suyu et al. 2012) will likely significantly increase this. Nevertheless, it is interesting that the LAT time delay brings the estimated value of Hubble's constant down, toward the low end of modern measurements (e.g., Planck Collaboration 2013). An underdense environment would require this inferred h value to increase; including external lensing effects in future cosmographic analyses might be important in this system. Moreover, since the radio and γ-ray emission regions are likely not co-spatial, the assumed radio-derived time-delay function values may be inaccurate. A fully self-consistent joint modeling of the radio and γ-ray source is needed to resolve this. If the LAT can measure a lag in the γ-ray light curve of one of the previously known systems with wider separation or in a new example (below), this can give independent γ-ray based constraints on Hubble's constant.

One exciting result would be the detection of a lens delay in a flaring γ-ray source that is not yet identified as a gravitationally lensed system at radio wavelengths or otherwise. These could possibly be lensed image pairs with flat-spectrum radio sources at smaller separations than in the 02 resolution of VLA surveys (references above). Similar radio lens surveys in the southern hemisphere are not yet as complete (e.g., Prouton et al. 2001), so a γ-ray delay signature in their LAT light curves could betray the presence of a previously unknown lens system. Such a strategy has been proposed for future wide-field optical surveys (Pindor 2005), and the discovery potential of the LAT in γ-rays should now be recognized. Furthermore, with the different flux ratios at radio and γ-ray wavelengths, and possible variability of the ratio, some sources could be bright in γ-rays and less conspicuous at radio. Such potential gravitational lenses could be hidden in plain sight within the radio catalogs used for blazar associations in LAT catalogs, or could be among the currently unidentified γ-ray sources (Torres et al. 2002).

The Fermi-LAT Collaboration acknowledges support from a number of agencies and institutes for both development and the operation of the LAT as well as scientific data analysis. These include NASA and DOE in the United States, CEA/Irfu and IN2P3/CNRS in France, ASI and INFN in Italy, MEXT, KEK, and JAXA in Japan, and the K.A. Wallenberg Foundation, the Swedish Research Council, and the National Space Board in Sweden. Additional support from INAF in Italy and CNES in France for science analysis during the operations phase is also gratefully acknowledged. C.C.C. was supported at NRL in part by NASA DPR S-15633-Y.

Facility: Fermi - Fermi Gamma-Ray Space Telescope (formerly GLAST)