Spitzer Observations of the Predicted Eddington Flare from Blazar OJ 287

Visibility and scheduling constraints did not permit Spitzer (Werner et al. 2004) to observe OJ 287 until 2019 July 31, 15 UT, several hours after the predicted time window for the occurrence of the impact flare peak. Therefore, we focused on the declining part of the expected flare where the radiating bubbles are optically thin in all relevant wavebands (Valtonen et al. 2019). This part lies between the first brightness peak and the first major minimum, predicted to occur during 2019 August 7. The Spitzer scheduling permitted dense monitoring during this critical period. Altogether OJ 287 was observed with Spitzer's Infrared Array Camera (Fazio et al. 2004) on 21 epochs between 2019 July 31 and 2019 September 6. The cadence was approximately 12 hr for the first five epochs, then once per day for the next eight epochs, and thereafter approximately twice a week for the last eight epochs. Additionally, OJ 287 was monitored on five epochs between 2019 February 25 and 2019 March 2 with daily cadence for normalization purposes with simultaneous optical observations. February–March observations permitted us to convert the infrared Spitzer/IRAC channel-1 (3.6 μm) and channel-2 (4.5 μm) flux densities, observed during the flare, to equivalent R-band flux densities. These observations were taken as part of the Spitzer DDT program pid 14206. The observing log and reduced flux densities are given in Table 1.

All of the observations were taken both in the 3.6 and 4.5 μm channels (corresponding approximately to the conventional photometric L- and M-bands) using the 2 s frame time with a 10-position medium-scale dither (typical dither amplitudes of less than an arcminute). The same dither starting point was used in every observation so that OJ 287 landed roughly on the same pixel position in each observation and dither offset.

The corrected basic calibrated data frames were inspected by eye, and remaining artifacts, such as column pulldown, were removed with the imclean tool.²⁹ Frames with a cosmic-ray detection within a 10-pixel (approximately 12'') radius of OJ 287 were not included in the analysis (in general there were zero to one such frames per observation). The centroid of the image of OJ 287 was found with the first moment centroiding method.³⁰ We performed aperture photometry with the IDL procedure aper using a source aperture radius of six pixels and a background annulus between 12 and 20 pixel radial distance from the centroid position. We corrected the flux densities with the irac_aphot_corr.pro procedure,³¹ for the pixel phase and array location-dependent response functions. In addition, we performed an aperture correction as tabulated in the IRAC Instrument Handbook.³² For each channel, at each epoch, we finally calculated the mean flux density and the uncertainty from the standard error of the mean, as presented in Table 1.

Recall that we predicted the optical R-band lightcurve for the 2019 impact flare from the corresponding observations in 2007. However, our observations of the Eddington flare are in the two near-infrared Spitzer channels. Therefore, it is crucial to estimate how the predicted flare lightcurve should look in the Spitzer bands. In the quiescent state, the infrared–optical wavelength emission comes from synchrotron radiation, and the spectrum follows a power law with a spectral index α ∼ −0.95 (Kidger et al. 2018). In contrast, BH impact flares are dominated by bremsstrahlung radiation, which has a nearly flat spectrum in the near-infrared–optical wavelengths superposed on the usual synchrotron emission. Therefore, we expect that the impact-induced fluxes in the Spitzer bands will be similar to those in the optical bands. However, the base levels of the fluxes in the Spitzer and optical wavelengths should be different during such outbursts due to the steep power-law spectrum of the synchrotron background. Therefore, we subtract the base-level fluxes from the observed Spitzer band fluxes during the outburst to compare with the predicted R-band flux curve. We expect the 2019 impact flare to be coincident in time in the optical and the near-infrared as multiwavelength observations of the 2015 impact flare show no time delay across the relevant wavebands (Valtonen et al. 2016; Kushwaha et al. 2018).

We now examine if our observed Spitzer lightcurve does contain the predicted impact flare. This requires us to create a template of the expected flare, as given in Figure 2, and we focus on the declining part after the first peak which lasts around 7 days. The template is obtained by fitting a polynomial to the observed R-band lightcurve of the 2007 flare (Valtonen & Sillanpää 2011), shifted forward by 11.8752 yr. This time shift between the 2007 and 2019 flares, with ±4 hr uncertainty, was previously found in the orbit solution (D18). We introduce the three parameters Δt, ΔF₁, and ΔF₂ for fitting the Spitzer data with the outburst template. The parameters ΔF₁ and ΔF₂ are used to correct for the expected base-level differences between R-band and Spitzer's Ch-1 and Ch-2 fluxes, respectively. The Δt parameter allows us to find the difference between the predicted and actual arrival times of the 2019 outburst. Note that it shifts the time variable in our polynomial fit for the 2007 lightcurve. We employ only a single Δt parameter for both channels as we expect the impact flare to produce simultaneous flux variations in both channels. The best-fit values with 1σ uncertainties read Δt = −0.06 ± 0.05 days, ΔF₁ = 13.92 ± 0.11 mJy, and ΔF₂ = 19.55 ± 0.09 mJy. This implies that the Eddington flare arrived 1.4 ± 1.2 hr late of the predicted epoch but well within the expected time interval. Therefore, we shift our flux templates forward in time by 0.06 days to obtain Figure 2 where we compare the base-level corrected flux variations in Spitzer channels with the template of 2007. We also performed a self-consistency test by fitting the Ch-1 and Ch-2 fluxes separately, and the resulting values of Δt's for the two channels agree with each other within their uncertainties, as required. Qualitatively, the predicted lightcurve template for the 2019 impact outburst matches fairly well with the base-level corrected fluxes of both Spitzer channels. To quantify these similarities, we computed Pearson's r between the observed Spitzer data sets and the time-corrected template of Figure 2 and found high correlations (Pearson's r ∼ 0.98). We repeated this analysis using 20,000 random 1 week long OJ 287 lightcurves to rule out the occurrence of high Pearson's r values due to chance coincidences.

Zoom In Zoom Out Reset image size

It turns out that the possible template choices introduce ∼1 hr uncertainty in the flare timing. The template curve of Figure 2 should actually be a Gaussian band instead of a single line, since there is always some background noise in the source, and because the 2007 observations have associated error bars. Instead of a single template curve there could be any number of alternative ones that fit inside the band of ±0.3 mJy vertical half-width. Repeating the above processes with this band instead of a single line widens the error bars in ΔF but has no effect on Δt beyond the 1 hr additional error. Further, the radiating bubble that emits bremsstrahlung with a Maxwellian velocity distribution can give the spectral index α ∼ −0.2, rather than 0.0, the exactly flat spectrum (intensity ∼ ν^α) if the source has a constant temperature T (Karzas & Latter 1961). However, when we are looking at an expanding bubble, the light travel time is different from different parts of the source, and therefore the spectrum is composed of contributions from different temperatures T within some range ΔT. The intensity depends essentially only on the parameter u = hν/kT. If we employ a reasonable assumption that temperatures T are uniformly distributed over this range, then the intensity is constant over the corresponding range in frequency Δν and we get a flat emission spectrum. Naturally, details depend on the models of the emitting bubble (Pihajoki 2016). It is likely that the spectral index α lies between −0.2 and 0.0. For α = −0.2, we get Δt = 0.10 ± 0.05 implying that the flare arrived 2.5 ± 1.2 hr early. Thus, considering these uncertainties, the Eddington flare came within ∼4 hr of the predicted time.

The nearly flat spectrum of the impact flare should cause an overall decrease in the ratio of 4.5 and 3.6 μm fluxes in the neighborhood of flare peak in Spitzer data. Plots in Figure 3 confirm this expectation. The flux ratios during the outburst window from 2019 July 31 to August 6 have a significantly different distribution with smaller values compared to their counterparts during the non-outburst phases. Further, we got a Kolmogorov–Smirnov (K-S) statistic of 0.66 with a p-value = 0.0053 while doing the K-S test between the two set flux ratios. At this significance level, the flux ratios during the outburst and non-outburst epochs come from two distinct distributions.

Zoom In Zoom Out Reset image size

It is also possible to construct a fiducial R-band magnitude flare lightcurve from Spitzer data using the calibration measurements that involved both optical R-band and the Spitzer channels during 2019 February 25–2019 March 2. We find that the Ch-1 flux can be converted to an equivalent optical R-band value by dividing the former by ∼6.2 and for Ch-2 the factor is ∼7.6. Therefore, using the previously obtained base-level contributions to the infrared flux, the R-band base flux during the 2019 impact flare should be ∼3.73 mJy. Thereafter, this R-band flux is added to the excess flux above the base level in the two Spitzer bands. The resulting two fluxes are averaged at every epoch and converted to R magnitudes (using the Gemini observatory converter). This is plotted in Figure 4, together with the actual ground-based R-band observations. Indeed, the fiducial R-band magnitudes join smoothly with the direct R-band observations in early September where both Spitzer and optical observations overlap. These plots endorse the similar nature of 2007 and 2019 flares both in their total sizes and general lightcurve shapes.

Zoom In Zoom Out Reset image size