MULTIWAVELENGTH OBSERVATIONS OF 3C 454.3. III. EIGHTEEN MONTHS OF AGILE MONITORING OF THE "CRAZY DIAMOND"

The AGILE satellite (Tavani et al. 2008, 2009), a mission of the Italian Space Agency (ASI) devoted to high-energy astrophysics, is currently the only space mission capable of observing cosmic sources simultaneously in the energy bands 18–60 keV and 30 MeV–50 GeV.

The AGILE scientific instrument combines four active detectors yielding broadband coverage from hard X-rays to γ-rays: a Silicon Tracker (ST; Prest et al. 2003, 30 MeV–50 GeV), a co-aligned coded-mask hard X-ray imager, Super-AGILE (SA; Feroci et al. 2007, 18–60 keV), a non-imaging CsI Mini-Calorimeter (MCAL; Labanti et al. 2009, 0.3–100 MeV), and a segmented Anti-Coincidence System (ACS; Perotti et al. 2006). Gamma-ray detection is obtained by the combination of ST, MCAL, and ACS; these three detectors form the AGILE Gamma-Ray Imaging Detector (GRID).

Level-1 AGILE-GRID data were analyzed using the AGILE Standard Analysis Pipeline (see V08 for a detailed description of the AGILE data reduction). Since 3C 454.3 was observed at high off-axis angle due to the satellite solar panel constraints, an ad hoc γ-ray analysis was performed. We used γ-ray events filtered by means of the FM3.119_2a AGILE Filter pipeline. Counts, exposure, and Galactic background γ-ray maps were created with a bin size of 025 × 025 , for E ⩾ 100 MeV. Since the source was observed up to 40° off-axis, all the maps were generated including all events collected up to 60° off-axis. We rejected all γ-ray events whose reconstructed directions form angles with the satellite-Earth vector smaller than 85°, reducing the γ-ray Earth albedo contamination by excluding regions within ∼15° from the Earth limb. We used the version (BUILD-16) of the Calibration files (I0006), which are publicly available at the ASI Science Data Centre (ASDC) Web site, ⁵³ and of the γ-ray diffuse emission model (Giuliani et al. 2004). We ran the AGILE Source Location task in order to derive the most accurate location of the source. Then, we ran the AGILE Maximum Likelihood Analysis (ALIKE) using a radius of analysis of 10°, and the best guess position derived in the first step. The particular choice of the analysis radius is dictated to avoid any possible contamination from very off-axis residual particle events.

Table 1 shows the AGILE/GRID observation log during the different time periods. We have re-analyzed all the AGILE data already published in V08, Paper I, and Paper II, respectively, according to the procedure described in Section 2.1. The results are discussed in Section 7. In the following paragraphs we report the detailed results of the still unpublished AGILE data.

2.2.1. 2008 May–June

The AGILE campaign was split into two different periods, May 10–June 9 (P1) and June 15–30 (P2) because of a ToO re-pointing toward W Comae. The total on-source exposure is 1.03 (P1) + 0.54 (P2) Ms. 3C 454.3 was detected, during P1 and P2, at a 25.6σ and 16.3σ level with an average flux of F^P1 _E>100 MeV = (218 ± 12) × 10⁻⁸ photons cm⁻² s⁻¹, and F^P2 _E>100 MeV = (198 ± 17) × 10⁻⁸ photons cm⁻² s⁻¹ , respectively, as derived from the AGILE Maximum Likelihood Code analysis.

Figure 1, filled circles in panel (a), shows the γ-ray light curve at 1 day resolution for photons above 100 MeV. We note that, particularly at the beginning of the campaign, 3C 454.3 was not always detected on a day-by-day timescale. On MJD ∼ 54610 the source began to be detected at a 3σ level almost continuously; this clearly indicates the onset of a γ-ray flaring activity.

Zoom In Zoom Out Reset image size

The average γ-ray flux as well as the daily values were derived according to the γ-ray analysis procedure described in Mattox et al. (1993). First, the entire period was analyzed to determine the diffuse gas parameters and then the source flux density was estimated independently for each of the eighteen 1 day periods with the diffuse parameters fixed at the values previously obtained.

Figure 1, panel (b), shows the same AGILE/GRID data binned on a timescale of 3 days. The light curve clearly shows a strong degree of variability, with a dynamic range of about four in about two weeks.

Figure 2, panel (a), shows the average γ-ray spectra extracted over the observing periods P1 and P2. Each average spectrum was obtained by computing the γ-ray flux in five energy bins over each period: 50 MeV < E < 100 MeV, 100 MeV < E < 200 MeV, 200 MeV < E < 400 MeV, 400 MeV < E < 1000 MeV, and 1000 MeV < E < 3000 MeV. We note that the current instrument response is accurately calibrated in the energy band 100 MeV–1 GeV, and that the flux above 1 GeV is underestimated by a factor of about 2. For those reasons, we fit the data by means of a simple power-law model and restricted our fit to the 100 MeV–1 GeV energy range, obtaining

$$\begin{eqnarray} F^{\rm P1}(E) &=& 2.63 \times 10^{-4} \times \left(\frac{{ E}}{1\, {\rm MeV}}\right)^{-(2.05 \pm 0.10)}\nonumber\\ &&\qquad{\rm photons\;cm^{-2} s^{-1}\, MeV^{-1}}, \end{eqnarray} \tag{ 1 }$$

$$\begin{eqnarray} F^{\rm P2}(E) &=& 1.58 \times 10^{-4} \times \left(\frac{{ E}}{1\, {\rm MeV}}\right)^{-(1.98 \pm 0.16)}\nonumber\\ &&\qquad{\rm photons\;cm^{-2} s^{-1} \, MeV^{-1}}. \end{eqnarray} \tag{ 2 }$$

Zoom In Zoom Out Reset image size

The different energy range, and, above all, the different time period, could explain the different value of the AGILE (2.05 ± 0.10 and 1.98 ± 0.16) and Fermi/LAT (2.27 ± 0.03 , pre-break; Abdo et al. 2009) γ-ray photon indices.

2.2.2. 2008 July–August

The AGILE campaign started immediately after the Fermi/LAT detection of a very high γ-ray activity in the period 2008 July 10–21 (Tosti et al. 2008), which reached, on July 10, a γ-ray flux of F^Fermi _E>100 MeV = 1200 × 10⁻⁸ photons cm⁻² s⁻¹  (Abdo et al. 2009). The AGILE observations covered the period from 2008 July 25 19:57 UT to 2008 August 14 21:08 UT, for a total on-source exposure of about 0.71 Ms. 3C 454.3 was detected at a 17.5σ level with an average flux of F^ja08 _E>100 MeV = (255 ± 21) × 10⁻⁸ photons cm⁻² s⁻¹, as derived from the AGILE Maximum Likelihood Code analysis.

Figure 1, filled squares in panel (a) and in panel (b), shows the γ-ray light curve at 1 day and at 3 day resolution, respectively, for photons above 100 MeV. The average γ-ray flux as well as the daily values were derived according to the procedure described in Section 2.2.1. Contrary to the May–June data, the July–August light curve does not show any clear sign of variability.

Figure 2, panel (b), shows the average γ-ray spectrum derived over the entire observing period. The average spectrum was obtained by computing the γ-ray flux in the same way as in Section 2.2.1. We fit the data by means of a simple power-law model and restricted our fit to the most reliable energy range (100 MeV–1 GeV):

$$\begin{eqnarray} F^{\rm ja08}(E) &=& 3.96 \times 10^{-4} \times \left(\frac{{ E}}{1\, {\rm MeV}}\right)^{-(2.11 \pm 0.14)}\nonumber\\ &&\qquad{\rm photons\;cm^{-2} s^{-1}\, MeV^{-1}}. \end{eqnarray} \tag{ 3 }$$

During this period, which partially overlaps with the Fermi one, the AGILE γ-ray photon index is, within the statistical errors, in agreement with the Fermi/LAT result.

2.2.3. 2008 October–2009 January

The AGILE observations covered the period from 2008 October 17 12:51 UT to 2009 January 12 14:30 UT, for a total on-source exposure of about 2.86 Ms. 3C 454.3 was detected at a 17.9σ level with an average flux of F^oj09 _{E > 100 MeV} = (77 ± 5) × 10⁻⁸ photons cm⁻² s⁻¹, as derived from the AGILE Maximum Likelihood Code analysis.

Figure 1, filled triangles in panel (a), shows the γ-ray light curve at 1 day resolution for photons above 100 MeV. The average γ-ray flux as well as the daily values were derived according to the procedure described in Section 2.2.1. The light curve does not show any clear trend, partly because of a dominant fraction of upper limits in the data. For this reason, we decided to rebin the light curve on a timescale of one week. The resulting light curve is shown in Figure 1, filled triangles in panel (b). On this timescale, a clear trend is present, with the source dimming as a function of time, with a dynamic range of about a factor of 2.

Figure 2, panel (c), shows the average γ-ray spectrum derived over the entire observing period. The average spectrum was obtained by computing the γ-ray flux in the same way as in Section 2.2.1. We fit the data by means of a simple power-law model and restricted our fit to the most reliable energy range (100 MeV–1 GeV):

$$\begin{eqnarray} F^{\rm oj09}(E) &=& 2.10 \times 10^{-4} \times \left(\frac{{ E}}{1\, {\rm MeV}}\right)^{-(2.21 \pm 0.13)}\nonumber\\ &&\qquad{\rm photons\;cm^{-2} \, s^{-1} \; MeV^{-1}}. \end{eqnarray} \tag{ 4 }$$

During the various AGILE pointings, 3C 454.3 was located substantially off-axis in the Super-AGILE field of view (FoV). For this reason, only 3σ upper limits can be derived in the 20–60 keV energy band during the AGILE/GRID observations. Table 2 summarizes the Super-AGILE observations and their results.

Swift pointed observations (Gehrels et al. 2004) were performed from 2007 July 26 to 2009 January 1. These observations were obtained both by means of several dedicated ToOs (P.I.: S. Vercellone) and by activating Swift Cycle-3 (Obs. ID 00031018, P.I.: A.W. Chen) and Cycle-4 Proposals (Obs. ID 00031216, P.I.: S. Vercellone). A long-lasting monitoring program (P.I.s: L. Fuhrmann and S. Vercellone) covers the period 2008 July–October.

3.1.1.  Swift/XRT

3.1.2.  Swift/UVOT

3.1.3.  Swift/BAT

Figure 3 shows the Swift/XRT fluxes in the 2–10 keV energy range, the Swift/UVOT observed magnitudes (in the V, B, U, W1, M2, and W2 bands), and the Swift/BAT fluxes in the 15–150 keV energy range as a function of time for the whole observing period. In order to diminish the statistical uncertainties, we selected observations with a number of degrees of freedom (dof) >10. We note that a common dimming trend is present both in the UV and in the X-ray energy bands.

Zoom In Zoom Out Reset image size

As shown in Figure 3, panel (d), the source has not been always detectable by Swift/BAT throughout the considered period, and in several time intervals only 3σ upper limits can be derived.

Figure 4 shows the Swift/XRT photon index as a function of the X-ray flux in the 2–10 keV energy band. Black circles and red squares represent data acquired in PC and WT mode, respectively. WT data are not affected by pile-up at the observed count rate (CR < 3 counts s⁻¹). We investigated the possible presence of a spectral trend in the X-ray data. If we consider WT data only, a "harder-when-brighter" trend seems to be present. Fitting the data with a constant model, we can exclude this model at the 99.9993% level. When analyzing the PC data only (as well as the sum of the PC and WT data), this spectral trend vanishes, and a fit with a constant model still holds. Nevertheless, if we exclude the points at fluxes F_2-10 keV < 2 × 10⁻¹¹ erg cm⁻² s⁻¹ a trend still holds. These points at low fluxes could correspond to physically different state of the source than the high fluxes one.

Zoom In Zoom Out Reset image size

We also note that a deep and prolonged monitoring of 3C 454.3 at mid and low X-ray states (F_2-10 keV ≲ 10⁻¹¹ erg cm⁻² s⁻¹) will be crucial to test the possible presence of a spectral trend. Our data set contains only four observations (90023002, 90023003, 90023006, and 90023008) at this flux level, which were acquired during the source low state in 2008 December.

The RXTE satellite observed 3C 454.3 in two epochs: from 2007 July 28 to 2007 August 04 and from 2008 May 30 to 2008 June 19. Here we report the analysis of the data obtained both with the Proportional Counter Array (PCA; Jahoda et al. 1996), which is sensitive in the 2–60 keV energy range, and with the High-Energy X-Ray Timing Experiment (HEXTE; Rothschild et al. 1998), which is sensitive in the 15–250 keV energy range. RXTE data were collected by activating a Cycle 12 ToO proposal (ID. 93150, P.I.: A.W. Chen).

The PCA is composed of five identical units (Proportional Counter Units, PCUs), but during our observations only part of them were used. Since PCU2 was the only unit always on during our observations and it is the one which is best calibrated, we report the results obtained from the PCU2 data only. The data were processed using the FTOOLS ver. 6.4.1 and screened using standard filtering criteria. The net exposure times for the whole data set in the first and second epochs were 36.6 ks and 17.4 ks, respectively.

The background light curves and spectra for each observation were produced using the model appropriate to faint sources. We restricted our analysis to the 3–20 keV energy range, in order to minimize the systematic errors due to background subtraction and calibration of the PCA instrument.

Figure 5 shows the 3–20 keV light curve of the whole RXTE/PCA data set. Strong variability is observed when comparing the count rates of different observations. Moreover, the average count rate during the second epoch (panel (b)) is reduced. In order to investigate possible changes in the spectral shape with time we extracted light curves in two energy ranges (3–7 keV and 7–20 keV). Their hardness ratio did not show any significant variation.

Zoom In Zoom Out Reset image size

A cumulative spectrum for the first and the second epochs was extracted and simultaneously fitted with a power-law model corrected for photoelectric absorption (wabs*(powerlaw) model in XSPEC), allowing only the power-law normalization to assume a different value in the two spectra. Figure 6 shows the RXTE/PCA spectra for both periods. A good fit (χ²= 68.3 for 76 dof) was obtained with the following best-fit parameters (errors are at the 90% confidence level): photon index Γ = 1.65 ± 0.02, and a flux in the 3–20 keV energy band F_3-20 keV = 8.4 × 10⁻¹¹ erg cm⁻² s⁻¹ and F_3-20 keV = 4.5 × 10⁻¹¹ erg cm⁻² s⁻¹ for the first and second epoch spectrum, respectively.

Zoom In Zoom Out Reset image size

We note that the average RXTE/PCA flux during the γ-ray flare detected in 2007 July was about a factor of 2 higher than the flux detected about 10 months later. Moreover, during both the 2007 July and the 2008 May–June campaigns, the hard X-ray flux varied significantly, by about 50%, on a timescale of about one week.

We also analyzed the data of the HEXTE. Only the data from cluster B were analyzed, since the rocking system for background evaluation was disabled in the other instrument cluster. After a standard processing, ⁵⁵ we extracted an average spectrum from all the available data, for a dead-time corrected exposure time of 18 ks. The source was detected up to ∼50 keV and its spectrum can be fit well (χ²=15.1 for 19 dof) by a power-law model with a photon index of 1.6 ± 0.1, in perfect agreement with the photon index derived from the PCA spectrum. Also, the normalization of the HEXTE spectrum is consistent with a high-energy extrapolation of the time-averaged PCA spectrum: the observed flux in the 20–40 keV energy band is (5 ± 3)×10⁻¹¹ erg cm⁻² s⁻¹. This flux (approximately 6 mCrab) is also consistent with the upper limits obtained by Super-AGILE in the same time periods.

The WEBT ⁵⁶ has been monitoring 3C 454.3 since the exceptional 2004–2005 outburst (Villata et al. 2006, 2007; Raiteri et al. 2007, 2008a, 2008b), throughout the whole period of the AGILE observation. We refer to Raiteri et al. (2008a, 2008b) and to Villata et al. (2009) for a detailed presentation and discussion of the radio, mm, near-IR, optical, and Swift/UVOT data.

Figure 7 shows the GASP-WEBT light curve in the R optical band, displaying several intense flares with a dynamic range of ∼2.4 mag in about 14 days, while Figure 8 shows the GASP-WEBT light curves in the near-IR (J, H, K), radio (5, 8, and 14.5 GHz), and mm (37, 230, and 345 GHz), respectively. ⁵⁷

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

High-resolution radio VLBI data were obtained from the MOJAVE (Monitoring Of Jets in Active galactic nuclei with VLBA Experiments) project, a long-term program to monitor radio brightness and polarization variations in jets associated with active galaxies visible in the northern sky (Lister et al. 2009; see also ⁵⁸ ). The object was observed with the full VLBA at 15 GHz. We obtained the calibrated I images and used the AIPS package to derive the position and flux density of the core and of a few substructures in the jets using the task JMFIT (Gaussian fit; see Figure 9). Moreover, this source was additionally observed by VLBA at four epochs during the period of the maximum brightness within the BK150 VLBA experiment to measure parsec-scale spectra of γ-ray bright blazars (K. Sokolovsky et al. 2010, in preparation). We use 15 and 43 GHz results from this program to provide better radio coverage of the high activity period. These data are in agreement with MOJAVE results and give a better statistics in the high active period.

Zoom In Zoom Out Reset image size

The core is always unresolved by our Gaussian fit and uncertainties on the flux density are dominated by calibration uncertainties (a few percent).

In Figure 10, we show the 3C 454.3 VLBI radio core flux (panel (a)) at 15 and 43 GHz, the radio components flux density at 15 GHz (panel (b)), and the distance of radio components from the core (panel (c)) as a function of time.

Zoom In Zoom Out Reset image size

The flux shows a constant increase from 2006 June 15 (MJD 53901) till 2008 October 3 (MJD 54742), followed by a fast decrease toward the last epoch presented here, 2009 June 25 (MJD 55007). Jet components show a well defined flux density decrease (component 1) or a slower flux density decrease which becomes almost constant in the last epochs. Proper motion is evident, but slowing in time for components 1 and 2; it is almost absent for component 3.

All data are in agreement with a strong core flux density variability possibly connected to the γ-ray activity, while jet components are moving away and slowly decreasing in flux density, and are not affected by the recent core activity. In a recent paper, Kovalev et al. (2009) indeed find a connection between the radio and the γ-ray emission, correlating the Fermi three month data with the MOJAVE ones, and arguing that the central region of the blazars being the source of γ-ray flares. Nevertheless, a detailed study of the radio structure of 3C 454.3 is beyond the aims of this paper, therefore the jet properties will be discussed in depth elsewhere (M. L. Lister et al. 2010, in preparation; S. G. Jorstad et al. 2010, in preparation). For this reason in the following we will concentrate only on the core.

In the last two years this source has also been observed with the VLBA at 43 GHz (S. G. Jorstad et al. 2010, in preparation; see also the Boston Univ. Blazar Group VLBA Web site ⁵⁹ ). We used the available images to derive the flux density of the core at 43 GHz. Note that, for a better comparison with 15 GHz VLBI data, at 43 GHz we used natural weights and we have not searched for possible core subcomponents (we refer to S. G. Jorstad et al. 2010, in preparation, for a detailed study of the radio structure).

The radio core shows an inverted spectrum (self-absorbed), more evident in the high active regime, followed by a strong flux density decrease. In this region the radio spectrum is no longer inverted.

The AGILE/GRID wide FoV allowed for the first time a long-term monitoring of 3C 454.3 at energies above 100 MeV. Moreover, coordinated and almost simultaneous GASP-WEBT and Swift observations provided invaluable information on the flux and spectral behavior from radio to X-rays.

Figure 12 shows the 3C 454.3 light curves at different energies over the whole period. The different panels show, from bottom to top, the AGILE/GRID light curve at ≈1 day resolution for E > 100 MeV in units of 10⁻⁸ photons cm⁻² s⁻¹, the Swift/BAT light curve in the energy range 15–150 keV at ≈2 week resolution, the Swift/XRT (filled circles) and the RXTE/PCA (filled squares) light curves in the energy range 3–10 keV, the Swift/UVOT light curve in the UV W2 filter, the Swift/UVOT light curve in the optical B filter, the GASP-WEBT light curve in the optical R filter, and the VLBI radio core at 15 GHz (filled circles) and the UMRAO 14.5 GHz (open circles) light curves, respectively.

Zoom In Zoom Out Reset image size

We note that RXTE/PCA data are systematically higher than Swift/XRT ones, which is consistent with the 20% uncertainty in the relative calibrations of the two instruments in this energy band, reported by Kirsch et al. (2005).

Figure 13 shows the light curves in the R band, at 1.3 mm (230 GHz), and above 100 MeV. The light curve in the millimeter wavelength shows a different behavior starting from the enhanced γ-ray activity at MJD ∼54600, as will be discussed in Section 8.5. Moreover, starting from MJD 54750, the whole jet seems to become less energetic, with an almost monotonic flux decrease, except for a minor burst at MJD 54800.

Zoom In Zoom Out Reset image size

We investigated the correlation between the γ-ray flux and the optical flux density in the R band by means of the discrete correlation function (DCF; Edelson & Krolik 1988; Hufnagel & Bregman 1992). This method was developed to study unevenly sampled data sets and can give an estimate of the accuracy of its results. Because of the sampling gaps in the light curves, especially at γ-rays, we calculated the DCF on four distinct periods: 2007 July (mid 2007), 2007 November–December (fall 2007), 2008 May–August (mid 2008), and 2007 October–2009 January (fall 2008). The upper limits on the γ-ray fluxes were considered as detections, with fluxes equal to one-half of the limit. In "mid 2007" AGILE was pointed at 3C 454.3 when its optical main peak was already over; furthermore, we only have five γ-ray points. The low statistics prevents us to obtain reliable results with the DCF for this period. In contrast, the period "fall 2007" offers a good opportunity to test the correlation, since the γ-ray flux, and even more the optical flux, exhibited strong variability. Moreover, the period of common monitoring lasted for more than a month. The corresponding DCF (Figure 14) shows a maximum DCF ∼ 0.38 for a null time lag. However, the shape of the peak is asymmetric, and if we calculate the centroid (Peterson et al. 1998), we find that the time lag is −0.42 days, i.e., about 10 hr. This result is in agreement with what was found by Donnarumma et al. (2009) when analyzing the 2007 December observations only, and implies that the γ-ray flux variations are delayed by few hours with respect to the optical ones. In the period "mid 2008" the main optical peak (and also the minor one) occurred when AGILE was not observing the source. Finally, we computed the DCF corresponding to the "fall 2008" period. We obtain a broad maximum, indicating a fair correlation (DCF_max ∼ 0.66), but with large errors, peaking at −2 day time lag, but with centroid around 0 day. This result is consistent with that obtained in the "fall 2007" period, which appears to be the most robust one. Hence, for this case we estimated the uncertainty on the time lag by means of the statistical method known as "flux randomization/random subset selection" (FR/RSS Peterson et al. 1998; Raiteri et al. 2003). We run 2000 FR/RSS realizations and for each of them calculated the centroid corresponding to the maximum. The resulting centroid distribution shown in Figure 14 allows us to conclude that the γ-optical correlation occurs with a time lag of τ = −0.4^+0.6 _−0.8, the uncertainty corresponding to a 1σ error for a normal distribution. This result is consistent with a recent analysis of the public Fermi data and the optical SMARTS data by Bonning et al. (2009).

Zoom In Zoom Out Reset image size

The observed variance of a light curve for a specific detector can be written as

$$\begin{equation} S^{2} = \frac{1}{N-1}\sum _{i=1}^{N} (x_{i} - \bar{x})^{2}, \end{equation} \tag{ 5 }$$

where $\bar{x}$ is the average value of the x_i measurements. Moreover, since we deal with different detectors, in order to take into account the different count rates in different energy bands, and to compare their variance, we consider the normalized variance, $S^{2}/\bar{x}^{2}$. In order to compute the intrinsic variance of a source light curve, the measurement errors must be taken into account, since they contribute an additional term to the variance. This approach was treated in detail by Nandra et al. (1997) and by Edelson et al. (2002), who introduced the term of "excess variance":

$$\begin{equation} \sigma _{\rm XS}^{2} = S^{2} - \bar{\sigma ^{2}}, \end{equation} \tag{ 6 }$$

where $\bar{\sigma ^{2}}$ is the mean squared error,

$$\begin{equation} \bar{\sigma ^{2}} = \frac{1}{N} \sum _{i=1}^{N} \sigma _{i}^{2}, \end{equation} \tag{ 7 }$$

and σ_i are the measurement uncertainties of light curve points x_i .

Thus, the normalized excess variance,

$$\begin{equation} \sigma _{\rm NXS}^{2} = \frac{\sigma _{\rm XS}^{2}}{\bar{x^{2}}}, \end{equation} \tag{ 8 }$$

can be used to compare variances between different observations.

In order to quantify the flux variability in different energy bands, we computed the fractional root mean square (rms) variability amplitude, F_var, defined as

$$\begin{equation} F_{\rm var} = \sqrt{\frac{S^{2} - \bar{\sigma ^{2}}}{\bar{x}^{2}} } \end{equation} \tag{ 9 }$$

(see also Vaughan et al. 2003, and references therein).

Figure 15 shows the fractional rms variability amplitude at different frequencies. The optical R band is the one showing the highest degree of variability. This is partly due to the higher sampling of the optical data with respect to the other frequencies. Nevertheless, a possible trend (higher fractional variability amplitude at higher frequency) is also present. This possible trend of the fractional rms variability amplitude with the logarithm of the frequency was observed in other sources too (see, e.g., PKS 2155 − 304, Zhang et al. 2005), and it was interpreted as signature of spectral variability. A more systematic study of the variability properties of 3C 454.3 will be addressed in a forthcoming paper.

Zoom In Zoom Out Reset image size

Figure 12 clearly shows a strong enhancement of the radio core flux starting about on MJD 54500. The highest flux density is on MJD 54742 at 15 GHz and on MJD 54719 at 43 GHz. This variability is not well correlated with the variability at higher frequencies: optical and γ-ray data show more different flares in the period MJD 54400–54800 (see Figure 13, panels (a) and (c), respectively). Moreover, the radio flux density increase is smooth and longer in time, while γ-ray and optical flares are evolving faster.

At 230 GHz the flux density variability mimics the VLBI radio core properties to MJD 54600, when a large flux density increase is visible, with a peak at about MJD 54630. At this frequency the source remains in an active phase up to MJD 54700 (Figure 13(b)).

This poses an interesting question as to the nature of such an increase of the core radio flux. As reported in Ghisellini et al. (2007) it is likely that the emitting region is more compact and has a smaller bulk Lorentz factor closer to the supermassive black hole. We can assume that in the region active at 43 GHz, in the quiescent state, the jet Lorentz factor is Γ ∼ 10 (Giovannini et al. 2001). To obtain the flux density increase of the core at 43 GHz (from ∼ 5 Jy up to 25 Jy) the Doppler factor has to increase up to δ ∼ 30. Such an increase requires that the source is oriented at a small angle θ with respect to the line of sight, since a large change in the jet velocity will produce a small increase in the Doppler factor. A Doppler factor δ = 30 can be obtained if θ = 15 and Γ = 20, corresponding to a bulk velocity increase from 0.9950 to 0.9987 (note that a larger orientation angle, e.g., θ = 3° with the same increase in the jet velocity, will produce a small change in the Doppler factor δ, from 16 to 19).

The presence of one or more new jet components is not revealed in the high resolution VLBA images, even if the most recent VLBA images at 43 GHz suggest a jet expansion near to the radio core starting from MJD ∼ 54600. Because of different properties (multiple bursts at high frequency, a single peak in the radio band) it is not possible to correlate the radio peak with a single γ-ray or optical burst. We can speculate that a multiple source activity in the optical and γ-ray bands is integrated in the radio emitting region in a single event. This event (see Figure 10) has a clear flux density peak on MJD ∼ 54720 and we can assume that 43 GHz is the self-absorption frequency at that epoch. This scenario is in agreement with the one discussed by Krichbaum et al. (2008, see their Figure 3).

According to Marscher (1983), the self-absorption frequency, the source size, flux density, and the magnetic field are correlated as follows:

$$\begin{equation} B = 3.2 \times 10^{-5} \times \theta ^{4}\, \nu _{\rm m}^{5}\, S_{\rm m}^{-2} \, \delta \, (1+z)^{-1} \,\, {\rm G}, \end{equation} \tag{ 10 }$$

where B is the magnetic field in G, θ is the angular size in mas (note that θ = 1.8 × HPBW, where HPBW is the half-power beam width), ν_m is the frequency (in GHz) of the maximum flux density S_m (in Jy), and δ is the Doppler factor, respectively. Moreover, we assume a particular value (α = 0.5) of spectral index in the optically thin part of the synchrotron spectrum.

Thus, we can use the radio VLBI data at 43 GHz to constrain the physical properties in the region where the source will start to be visible at this frequency. The angular resolution in the jet direction of VLBA data at 43 GHz is ∼ 0.14 mas corresponding (as discussed in Marscher 1983) to θ ≲ 0.25 mas. Assuming δ = 30, we obtain B ≲ 0.5 G.

It is reasonable to assume that when the source is even smaller, the emission in the radio band is not visible being self-absorbed, and that the local magnetic field is B ≲ 0.5 G when we start to detect the radio emission. The size of this region should be smaller than 0.25 mas (about 2 pc).

The correlation between the flux level and the spectral slope in the γ-ray energy band was extensively studied by means of the analysis of the EGRET data. Nandikotkur et al. (2007) showed that the behavior of EGRET blazars is inhomogeneous. Figure 16 shows the AGILE/GRID photon index as a function of the γ-ray flux at different epochs. A "harder-when-brighter" trend seems to be present in the long timescale AGILE data. Further long-term observations of 3C 454.3 and of other bright γ-ray blazars at different flux levels with AGILE and Fermi will be crucial to assess this topic.

Zoom In Zoom Out Reset image size

Different emission mechanisms can be invoked to explain the γ-ray emission. In the leptonic scenario, the low-frequency peak in the blazar SED is interpreted as synchrotron radiation from high-energy electrons in the relativistic jet, while the high-energy peak can be produced by IC on different kinds of seed photons. In the synchrotron self-Compton (SSC) model (Ghisellini et al. 1985; Bloom & Marscher 1996) the seed photons come from the jet itself. Alternatively, the seed photons can be those of the accretion disk (external Compton scattering of direct disk radiation, ECD; Dermer et al. 1992) or those of the BLR clouds (external Compton scattering from clouds, ECC; Sikora et al. 1994). The target seed photons can also be those produced by the dust torus surrounding the nucleus (external Compton scattering from IR-emitting dust, ERC(IR); Sikora et al. 2002).

We fit the SEDs for the different observing periods by means of a one-zone leptonic model, considering the contributions from SSC and from external seed photons originating both from the accretion disk and from the BLR (detailed description of this model is given in Vittorini et al. 2009). Indeed, emission from both of them were detected during faint states of the source (Raiteri et al. 2007).

The emission along the jet is assumed to be produced in a spherical blob with comoving radius R by accelerated electrons characterized by a comoving broken power-law energy density distribution of the form

$$\begin{equation} n_{e}(\gamma)=\frac{K\gamma _{\rm b}^{-1}}{(\gamma /\gamma _{\rm b})^{\alpha _{\rm l}}+ (\gamma /\gamma _{\rm b})^{{\alpha }_{\rm h}}}\,, \end{equation} \tag{ 11 }$$

where γ is the electron Lorentz factor assumed to vary between 10 < γ < 10⁴, α_l and α_h are the pre- and post-break electron distribution spectral indices, respectively, and γ_b is the break energy Lorentz factor. We assume that the blob contains a homogeneous and random magnetic field B and that it moves with a bulk Lorentz Factor Γ at an angle Θ₀ with respect to the line of sight. The relativistic Doppler factor is then δ = [Γ(1 − β cos Θ₀)]⁻¹, where β is the usual blob bulk speed in units of the speed of light.

Our modeling of the 3C 454.3 high-energy emission is based on an IC model with two main sources of external target photons: (1) an accretion disk characterized by a blackbody spectrum peaking in the UV with a bolometric luminosity L_d for an IC-scattering blob at a distance r_d = 4.6 × 10¹⁶ cm from the central part of the disk; and (2) a BLR with a spectrum peaking in the V band, placed at a distance from the blob of r_BLR = 4 × 10¹⁸ cm, and assumed to reprocess 10% of the irradiating continuum (Tavecchio & Ghisellini 2008; Raiteri et al. 2007, 2008b).

These two regions contribute to the ECD and the ECC, respectively, and it is interesting to test the relative importance of the two components that can be emitted by the relativistic jet of 3C 454.3 under different conditions. We summarize here the main results of our best model for the different time periods.

Table 5 shows the best-fit parameters of our modeling of SEDs corresponding to the following periods (see Figure 1): (SED1), MJD 54617–54618, when 3C 454.3 entered a phase of high γ-ray activity; (SED2) MJD 54673–54693, when the γ-ray flux was almost constant; (SED3) MJD 54800–54845, when the source flux was at the minimum level (about 70 × 10⁻⁸ photons cm⁻² s⁻¹ ). In Figures 17, 18, and 19, the thin solid, dotted, dashed, dot-dashed, and the triple-dot-dashed, represent the accretion disk blackbody, the synchrotron, the SSC, the external Compton on the disk, and the external Compton on the BLR radiation, respectively, while the thick solid line represents the sum of all the individual components. The inset of Figure 19 shows the portion of the SED dominated by the contribution of the disk blackbody radiation, which clearly emerges since the source is a relative low state.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We find that the three SEDs can be reproduced well by very similar parameters, the main difference being the shape of the electron distribution and the break energy Lorentz factor. We note that the observed minimum variability timescale, of the order of half a day, is consistent with the minimum variability timescale (∼10 hr) allowed by the model fit. Finally, we computed for the three different SEDs the total power carried in the jet, P_jet, defined as

$$\begin{equation} P_{\rm jet} = L_{\rm B} + L_{\rm p} + L_{\rm e} + L_{\rm rad}\,\,\,{\rm erg}\,{\rm s}^{-1}, \end{equation} \tag{ 12 }$$

where L_B, L_p, L_e, and L_rad are the power carried by the magnetic field, the cold protons, the relativistic electrons, and the produced radiation, respectively. We obtain a value of P_jet of 3.2 × 10⁴⁶ erg s⁻¹, 3.7 × 10⁴⁶ erg s⁻¹, and 2.5 × 10⁴⁶ erg s⁻¹ for SED1, SED2, and SED3, respectively. The total power of the jet is lower at the end of the AGILE observing period, following the general trend of the multiwavelength light curves.

The light curves in Figure 13 show a different behavior starting from the end of 2007 among the different energy bands.

As shown in Villata et al. (2009), a possible interpretation arises in the framework of a change in orientation of a curved jet, yielding different alignment configurations within the jet itself.

During 2007, the more pronounced fluxes and variability of the optical and γ-ray bands seem to favor the inner portion of the jet as the more beamed one. On the other hand, the dimming trend in the optical and in the γ-ray bands, the higher mm flux emission and its enhanced variability during 2008, seem to indicate that the more extended region of the jet became more aligned with respect to the observer line of sight.

The AGILE Mission is funded by the Italian Space Agency (ASI) with scientific and programmatic participation by the Italian Institute of Astrophysics (INAF) and the Italian Institute of Nuclear Physics (INFN). This investigation was carried out with partial support under ASI Contract No. I/089/06/1. We acknowledge financial support by the Italian Space Agency through contract ASI-INAF I/088/06/0 for the Study of High-Energy Astrophysics. This work is partly based on data taken and assembled by the WEBT collaboration and stored in the WEBT archive at the Osservatorio Astronomico di Torino - INAF. ⁶⁰ We thank the Swift and RXTE Teams for making these observations possible, particularly the duty scientists and science planners. This research has made use of data from the MOJAVE database that is maintained by the MOJAVE team. The 70 cm meniscus observations was partially supported by Georgian National Science Foundation grant GNSF/ST-08/4-404. E.K. acknowledges support from the NCS grant No. 96-2811-M-008-033. K. Sokolovsky was supported by the International Max Planck Research School (IMPRS) for Astronomy and Astrophysics at the Universities of Bonn and Cologne. We thank L. Fuhrmann for collaborating in the setup of the AGILE/Fermi Swift monitoring campaign in the period 2008 August–October. We also thank the referee for his/her constructive comments.

Facilities: AGILE - Italian Space Agency Astro-rivelatore Gamma ad Immagini LEggero satellite, Swift - Swift Gamma-Ray Burst Mission, RXTE - Rossi X-ray Timing Explorer, WEBT - Whole Earth Blazar Telescope, VLBA - Very Long Baseline Array, UMRAO - University of Michigan Radio Astronomy Observatory 26m telescope at Peach Mountain