RAPID VARIABILITY OF BLAZAR 3C 279 DURING FLARING STATES IN 2013−2014 WITH JOINT FERMI-LAT, NuSTAR, SWIFT, AND GROUND-BASED MULTI-WAVELENGTH OBSERVATIONS

The LAT is a pair-production telescope on board the Fermi satellite with large effective area ($\simeq 6500$ cm² on axis for 1 GeV photons) and a large field of view (2.4 sr), sensitive from 20 MeV to 300 GeV (Atwood et al. 2009). Here, we analyzed LAT data for the sky region including 3C 279 following the standard procedure²⁸ , using the LAT analysis software ScienceToolsv9r34v1 with the P7REP_SOURCE_V15 instrument response functions. The azimuthal dependence of the effective area was taken into account for analysis with short time scales (<1 day). Events in the energy range 0.1–300 GeV were extracted within a $15^\circ$ acceptance cone of the Region of Interest (ROI) on the location of 3C 279 (R.A. = $195\buildrel{\circ}\over{.} 047$, decl. = $-5\buildrel{\circ}\over{.} 789$, J2000). It is known that the Sun comes very close to and occults 3C 279 on October 8 each year. The data when the source is within $5^\circ$ of the Sun were excluded. Gamma-ray fluxes and spectra were determined by an unbinned maximum likelihood fit with gtlike. We examined the significance of the γ-ray signal from the sources by means of the test statistic (TS) based on the likelihood ratio test.²⁹ The background model included all known γ-ray sources within the ROI from the second Fermi-LAT catalog (2FGL: Nolan et al. 2012). Additionally, the model included the isotropic and Galactic diffuse emission components.³⁰ Flux normalizations for the diffuse and background sources were left free in the fitting procedure.

Several γ-ray light curves as measured by Fermi-LAT can be seen in Figure 1. The top stand-alone panel shows the γ-ray flux above 100 MeV for about 6 years since the beginning of scientific operations of the Fermi-LAT (2008 August 5) up to 2014 August 31 (MJD 54683–56900) binned into 3 day intervals. After ∼MJD 56600, the source entered the most active state since the launch of Fermi satellite. This resulted in target-of-opportunity (ToO) pointing observations for 3C 279, which were performed between 2014 March 31 21:59:47 UTC (MJD 56747.91652) and 2014 April 04 12:42:01 UTC (MJD 56751.52918), and those observations are included in our analysis. The time series of the γ-ray flux and photon index of 3C 279 measured with Fermi-LAT during the most active states from MJD 56615 (2013 November 19) to MJD 56775 (2014 April 28), are illustrated in other panels in Figure 1.

Zoom In Zoom Out Reset image size

Three distinct flaring intervals are evident in the γ-ray light curve: Flare 1 (∼MJD 56650), Flare 2 (∼MJD 56720) and Flare 3 (∼MJD 56750). The maximum 1 day averaged flux above 100 MeV reached $(62.2\pm 2.4)\times {10}^{-7}$ $\mathrm{photons}\ {\mathrm{cm}}^{-2}\ {{\rm{s}}}^{-1}$($\mathrm{TS}=3892$) on MJD 56749 (2014 April 03)³¹ , which is about three times higher than the maximum 1 day averaged flux recorded during the first two years (on MJD 54800: Hayashida et al. 2012). On the other hand, the maximum 1 day averaged flux above 1 GeV was observed on MJD 56645 (2013 December 20) at $(9.8\pm 1.2)\times {10}^{-7}$ $\mathrm{photons}\ {\mathrm{cm}}^{-2}\ {{\rm{s}}}^{-1}$, much higher than the $\gt 1$ GeV flux on MJD 56749, which was $(3.9\pm 0.4)\times {10}^{-7}$ $\mathrm{photons}\ {\mathrm{cm}}^{-2}\ {{\rm{s}}}^{-1}$. The photon index also shows a hardening trend toward MJD 56645, when it reached a very hard index of 1.82 ± 0.06, which is rarely observed in FSRQs.

Figure 2 shows detailed light curves around the flares with short time bins. During Flares 1 and 2, the fluxes were derived with an interval of 192 minutes, corresponding to two orbital periods of Fermi-LAT. During Flare 3, because the ToO pointing to 3C 279 increased the exposure, time bins as short as one orbital period (96 minutes) were used. The peak flux above 100 MeV in those time intervals (192 and 96 minutes) reached $\sim 120\times {10}^{-7}$ $\mathrm{photons}\ {\mathrm{cm}}^{-2}\ {{\rm{s}}}^{-1}$.

Zoom In Zoom Out Reset image size

The very rapid variability apparent in the data can be fitted by the following function to characterize the time profiles of the source flux variations:

$$\begin{eqnarray}&&F(t)={F}_{0}+\frac{b}{{e}^{-(t-{t}_{0})/{\tau }_{\mathrm{rise}}}+{e}^{(t-{t}_{0})/{\tau }_{\mathrm{fall}}}}.\end{eqnarray} \tag{ 1 }$$

This formula has also been used in variability studies of other LAT-detected bright blazars to characterize the temporal structure of γ-ray light curves (Abdo et al. 2010c). The double exponential form has been applied previously to the light curves of blazars (Valtaoja et al. 1999) as well as gamma-ray bursts(e.g., Norris et al. 2000). In this function, each ${\tau }_{\mathrm{rise}}$ and ${\tau }_{\mathrm{fall}}$ represents the "characteristic" time scale for the rising and falling parts of the light curve, respectively, and t₀ describes approximatively the time of the peak (it corresponds to the actual maximum only for symmetric flares). In general, the time of the maximum of a flare (t_p) can be described using parameters in Equation (1) as:

$$\begin{eqnarray}&&{t}_{p}={t}_{0}+\frac{{\tau }_{\mathrm{rise}}{\tau }_{\mathrm{fall}}}{{\tau }_{\mathrm{rise}}+{\tau }_{\mathrm{fall}}}\mathrm{ln}\left(\frac{{\tau }_{\mathrm{fall}}}{{\tau }_{\mathrm{rise}}}\right).\end{eqnarray} \tag{ 2 }$$

The parameters of the fitting results are summarized in Table 1. The time profiles show asymmetric structures in all flares; generally the rise times correspond to 1–2 hr, which are several times shorter than the fall times of 5–8 hr in Flares 1 and 3. On the other hand, the fall time appears to be less than 1 hour in Flare 2 (although the fitting error of the parameter is quite large). One can see in the light curve of Flare 2 in Figure 2 that the flux reached $\sim 90\times {10}^{-7}$ $\mathrm{photons}\ {\mathrm{cm}}^{-2}\ {{\rm{s}}}^{-1}$ at the peak but suddenly dropped by a factor of ∼3 in the next bin, two orbits (196 minutes) later.

Table 1.  Fitting Results of the Light Curve Profile in the γ-ray Band Measured by Fermi-LAT

Flare	${\tau }_{\mathrm{rise}}$	${\tau }_{\mathrm{fall}}$	b	F0	t0
Number	(hr)	(hr)	(10−7 $\mathrm{photons}\ {\mathrm{cm}}^{-2}\ {{\rm{s}}}^{-1}$)	(10−7 $\mathrm{photons}\ {\mathrm{cm}}^{-2}\ {{\rm{s}}}^{-1}$)	(MJD)
Flare 1	1.4 ± 0.8	7.4 ± 3.2	150 ± 36	19 ± 12	56646.35 ± 0.04
Flare 2	6.4 ± 2.4	0.68 ± 0.59	100 ± 26	19 ± 5	56718.32 ± 0.07
Flare 3 (ToO)	2.6 ± 0.6	5.0 ± 0.8	216 ± 19	10.5 ± 6.6	56750.30 ± 0.04

Download table as:  ASCIITypeset image

Gamma-ray spectra were extracted from the following four periods:

• 1.  
  (A) Overlapping with the first NuSTAR observation (see Section 2.2.1). Although the NuSTAR observation lasted for about one day, in order to increase the γ-ray photon statistics, the LAT spectrum was extracted from 3 days where the source showed comparable flux level (as inferred from the light curve with 1 day bins). In this period, the source was found to be in a relatively low state.
• 2.  
  (B) For three orbits (∼4.5 hr) at the peak of Flare 1, when the source showed a very hard γ-ray photon index (<2).
• 3.  
  (C) Overlapping with the second NuSTAR observation (see Section 2.2.1). As in the case of Period A, the length of this period is 3 days, while the NuSTAR observation lasted about 1 day. The source flux was higher than in Period A.
• 4.  
  (D) At the peak of Flare 3 for 4 orbits (∼6 hr).

In a similar manner to previous spectral studies of the source with the Fermi-LAT (Hayashida et al. 2012; Aleksić et al. 2014a), each γ-ray spectrum was modeled using a simple power-law (PL; ${dN}/{dE}\propto {E}^{-{{\rm{\Gamma }}}_{\gamma }}$), a broken power-law (BPL; ${dN}/{dE}\propto {E}^{-{{\rm{\Gamma }}}_{\gamma 1}}$ for $E\lt {E}_{\mathrm{brk}}$ and ${dN}/{dE}\propto {E}^{-{{\rm{\Gamma }}}_{\gamma 2}}$ otherwise), and a log-parabola model (LogP; ${dN}/{dE}\propto {(E/{E}_{0})}^{-\alpha -\beta \mathrm{log}(E/{E}_{0})}$, with ${E}_{0}=300$ MeV). The spectral fitting results are summarized in Table 2 and a spectral energy distribution for each period is plotted in Figure 3. In contrast to the general feature of FSRQs that the photon index is almost constant regardless of the source flux (see e.g., Hayashida et al. 2012), the spectral shape significantly changed between the periods. Remarkably, the photon index of the simple PL model for the Period B resulted in an unusually hard index for FSRQs, of 1.71 ± 0.10. Such a hard photon index has not been previously reported in past LAT observations of 3C 279 that included several flaring episodes (Hayashida et al. 2012; Aleksić et al. 2014a). Among the sources in the Second LAT Active Galactic Nucleus (AGN) Catalog (Abdo et al. 2010b), the mean photon index value of FSRQs is 2.4, and only one FSRQ (2FGL J0808.2−0750) in the clean and flux-limited sample has the photon index of ${{\rm{\Gamma }}}_{\gamma }\lt 2$ (see Figure 18 in Abdo et al. 2010b). Occasionally, hard photon indices have been observed in bright FSRQs during rapid flaring events (Pacciani et al. 2014). The photon index of Period B is even harder than the index of 4C + 21.35(1.95 ± 0.21; Aleksić et al. 2011b) and of PKS 1510−089(2.29 ± 0.02; Aleksić et al. 2014b) at the time when the >100 GeV emission was detected.

Zoom In Zoom Out Reset image size

Table 2.  Results of Spectral Fitting in the γ-ray Band Measured by Fermi-LAT

Period	Gamma-ray Spectrum (Fermi-LAT)						Flux ($\gt 0.1$ GeV)	# of Photons
(MJD—56000)	Fitting Modela	${{\rm{\Gamma }}}_{\gamma }/\alpha /{{\rm{\Gamma }}}_{\gamma 1}$	$\beta /{{\rm{\Gamma }}}_{\gamma 2}$	${E}_{\mathrm{brk}}$ (GeV)	TS	$-2{\rm{\Delta }}L$ b	$({10}^{-7}$ $\mathrm{photons}\ {\mathrm{cm}}^{-2}\ {{\rm{s}}}^{-1}$)	$\gt 10$ GeV
Period A (3 days)	PL	2.36 ± 0.13	⋯	⋯	174	⋯	5.9 ± 0.9	1
Dec 16, 0 h–19, 0 h	LogP	2.32 ± 0.17	0.03 ± 0.07	⋯	174	$\lt 0.1$	5.7 ± 0.9	(26.1 GeV)
(642.0–645.0)
Period B (0.2 days)	PL	1.71 ± 0.10	⋯	⋯	407	⋯	117.6 ± 19.7	1
Dec 20, 9h36–14h24	LogP	1.12 ± 0.31	0.19 ± 0.09	⋯	413	6.0	94.5 ± 18.1	(10.4 GeV)
(646.4–646.6)	BPL	1.41 ± 0.17	3.01 ± 0.91	3.6 ± 1.6	415	7.6	100.6 ± 18.4	⋯
Period C (3 days)	PL	2.29 ± 0.13	⋯	⋯	219	⋯	17.1 ± 2.8	1
Dec 31, 0 h–Jan 03, 0 h	LogP	2.29 ± 0.16	0.00 ± 0.06	⋯	219	$\lt 0.1$	17.1 ± 2.9	(14.7 GeV)
(657.0–660.0)	BPL	2.22 ± 0.42	2.32 ± 0.20	0.34 ± 0.27	219	$\lt 0.1$	16.9 ± 3.1	⋯
Period D (0.267 days)	PL	2.16 ± 0.06	⋯	⋯	1839	⋯	117.9 ± 7.1	1
Apr 03, 5h03–11h27	LogP	2.02 ± 0.08	0.10 ± 0.05	⋯	1840	5.3	114.9 ± 7.1	(13.5 GeV)
(750.210–750.477)	BPL	2.02 ± 0.09	2.89 ± 0.45	1.6 ± 0.6	1843	8.0	115.1 ± 7.7	⋯

^aPL: power law model, LogP: log-parabola model, BPL: broken-power-law model. See definitions in the text. ^b ${\rm{\Delta }}L$ represents the difference of the logarithm of the total likelihood of the fit with respect to the case with a PL for the source.

Download table as:  ASCIITypeset image

No significant deviations from a PL model were detected in the spectra of Periods A and C, while evidence of spectral curvature was observed in the spectra of the flare peaks, Periods B and D. As derived fitting the BPL model for Period B, the photon index of the lower energy part (below ${E}_{\mathrm{brk}}=3.6\pm 1.6$ GeV) is 1.41 ± 0.17. This is comparable to the photon index of the rising part of the inverse-Compton emission for the case of a parent electron index of 2, as is typical of the γ-ray spectra of high-frequency peaked BL Lac objects. One can easily recognize such a rising spectral feature in Figure 3. On the other hand, Period D also shows a very high flux, exceeding 10⁻⁵ $\mathrm{photons}\ {\mathrm{cm}}^{-2}\ {{\rm{s}}}^{-1}$ (>100 MeV), comparable to the flux of Period B. However, spectral shape is characterized by a soft index (${{\rm{\Gamma }}}_{\gamma }\gt 2$). The photon index of the lower energy part as derived from fitting with the BPL model is not significantly harder than 2, nor is a rising spectral feature apparent in the spectral energy distribution (SED) plot of Figure 3.

2.2.1. NuSTAR: Hard X-Rays

NuSTAR is a small explorer satellite sensitive to hard X-rays, covering the bandpass of 3–79 keV. It features two multilayer-coated optics, focusing the reflected X-rays onto CdZnTe pixel detectors which provide spectral resolution (FWHM) of 0.4 at 10 keV, increasing to 0.9 at 68 keV. The field of view of each telescope is $\sim 13\prime$, and the half-power diameter of the point spread function is $\sim 1\prime$. The low background resulting from focusing of X-rays provides an unprecedented sensitivity for measuring fluxes and spectra of celestial sources. For more details, see Harrison et al. (2013).

NuSTAR observed 3C 279 twice. The first observation was performed between 2013 December 16, 05:51:07 and 2013 December 17, 04:06:07 (UTC) and the second one between 2013 December 31, 23:46:07 and 2014 January 01, 22:11:07 (UTC). The raw data products were processed with the NuSTAR Data Analysis Software (NuSTARDAS) package v.1.3.1, using the nupipeline software module which produces calibrated and cleaned event files. We used the calibration files available in the NuSTAR CALDB calibration data base v.20140414. Source and background data were extracted from a region of $1\buildrel{\,\prime}\over{.} 5$ radius, centered respectively on the centroid of the X-ray source, and a region 5' N of the source location on the same chip. Spectra were binned in order to over-sample the instrumental resolution by at least a factor of 2.5 and to have a signal-to-noise ratio greater than 4 in each spectral channel. Net "on source" exposure times corresponded to ∼39.6 and ∼42.7 ks for the first (December 16) and second (December 31) observations, respectively. We considered the spectral channels corresponding nominally to the 3.0–70 keV energy range. The net (background subtracted) count rates for the first observation were 0.303 ± 0.003 and 0.294 ± 0.003 cnt s⁻¹ respectively for module A and module B, while for the second observation they were 0.636 ± 0.004 and 0.590 ± 0.004 cnt s⁻¹. We plotted the raw (not background subtracted) counts binned on an orbital time scale in Figure 4. It is apparent that the source was variable from one observation to the other, but also that the source varied within the second observation via secular decrease of flux for the first 3 hr, followed by an increase by nearly a factor of two.

Zoom In Zoom Out Reset image size

The spectral fitting was performed using XSPEC v12.8.1 with the standard instrumental response matrices and effective area files derived using the NuSTARDAS software module nuproducts. For each observation, we fitted the two modules simultaneously including a small normalization factor for module B with respect to the module A in the model parameters. We adopted simple PL and a BPL models modified by the effects of the Galactic absorption, corresponding to a column of $2.2\times {10}^{20}$ cm⁻²(Kalberla et al. 2005). The results of the two spectral fits were compared against each other by using an F-test to examine improvements by the BPL model. The simple PL model gave acceptable results for both observed spectra, with ${\chi }^{2}$/dof of 666.8/660 (41.9% for the corresponding ${\chi }^{2}$ probability) and 831.2/886 (90.6%), respectively. Although the model fluxes in the 2–10 keV band between the two observations showed a difference of about a factor of two, the resulting photon indices were similar: 1.739 ± 0.013, and 1.754 ± 0.008. While there was no improvement in the fit obtained using the BPL model in the first observation, it gave slightly better fits for the spectrum of the second observation, with a ${\chi }^{2}$/dof of 820.8/884 (93.6%), yielding a probability of 0.4% ($\sim 2.9\sigma$) that the improvement in the fit was due to chance (as assessed with an F-test). This may indicate a deviation from a single power law in the spectrum of the second observation. The fitting results for the NuSTAR spectra are summarized in Table 3.

Table 3.  Parameters of the Spectral Fits in X-ray Band

Instrument	${{\rm{\Gamma }}}_{{\rm{X}}}/{{\rm{\Gamma }}}_{{\rm{X}}1}$	${E}_{\mathrm{brk}1}$	${{\rm{\Gamma }}}_{{\rm{X}}2}$	${E}_{\mathrm{brk}2}$	${{\rm{\Gamma }}}_{{\rm{X}}3}$	Const.	${F}_{2-10\;\mathrm{keV}}$	${\chi }^{2}$/dof	F-test
(1)	(2)	(keV) (3)	(4)	(keV) (5)	(6)	XRT/module B (7)	(8)	(9)	(prob.)
Data on 2013 Dec 16–17 (in Period A)
Swift-XRT only	1.67 ± 0.08	⋯	⋯	⋯	⋯	⋯/⋯	11.9	21.10/18 (27.4%)	⋯
NuSTAR only	1.74 ± 0.01	⋯	⋯	⋯	⋯	⋯/1.06 ± 0.01	11.0	666.8/660 (41.9%)	⋯
XRT + NuSTAR	1.74 ± 0.01	⋯	⋯	⋯	⋯	$1.01\pm 0.05/1.06\pm 0.01$	11.0	688.6/679 (39.1%)	⋯
	${1.65}_{-0.08}^{+0.06}$	4.5 ± 0.7	1.75 ± 0.02	⋯	⋯	${1.11}_{-0.08}^{+0.09}/1.06\pm 0.01$	12.0	686.2/677 (39.5%)	30% (∼1.0$\sigma$)a
Data on 2013 Dec 31–2014 Jan 1 (in Period C)
Swift-XRT only	1.42 ± 0.03	⋯	⋯	⋯	⋯	⋯/⋯	23.7	113.1/109 (37.5%)	⋯
NuSTAR only	1.75 ± 0.01	⋯	⋯	⋯	⋯	⋯/1.00 ± 0.01	23.4	831.2/886 (90.6%)	⋯
	1.71 ± 0.02	${9.3}_{-1.3}^{+1.6}$	1.81 ± 0.02	⋯	⋯	⋯/1.00 ± 0.01	23.2	820.8/884 (93.6%)	0.4% (∼2.9$\sigma$)a
XRT + NuSTAR	1.73 ± 0.01	⋯	⋯	⋯	⋯	$0.70\pm 0.02/1.00\pm 0.01$	⋯	1072.4/996 (4.6%)	⋯
	1.37 ± 0.03	3.7 ± 0.2	1.76 ± 0.01	⋯	⋯	$0.97\pm 0.03/1.00\pm 0.01$	22.6	940.6/994 (88.6%)	⋯
	${1.37}_{-0.04}^{+0.03}$	${3.6}_{-0.4}^{+0.2}$	1.72 ± 0.02	${9.4}_{-1.4}^{+2.1}$	1.81 ± 0.02	0.97 ± 0.03/1.00 ± 0.01	22.6	933.4/992 (90.8%)	0.22% (∼3.1$\sigma$)b
(log parabola)c	$1.39\pm 0.03$ d	1 (fixed)e	$0.19\pm 0.02$ f	⋯	⋯	0.90 ± 0.03/1.00 ± 0.01	22.7	960.6/995 (77.8%)	⋯

Notes. Col. (1) instrument providing the data. Col. (2) photon index for the power law model, or low-energy photon index for the broken-power-law model. Col. (3) break energy (keV) for the broken-power-law model. Col. (4) high-energy photon index for the broken-power-law model. Col. (5) second break energy. Col (6) third index in the double-broken-power-law model. Col. (7) constant factor of Swift-XRT/NuSTAR module-B data with respect to the NuSTAR module-A data. Col. (8) unabsorbed model flux in the 2–10 keV band, in units of 10⁻¹² (erg cm⁻² s⁻¹). Col. (9): ${\chi }^{2}/$ degrees of freedom and a corresponding probability.

^aCompared to the simple-power-law model. ^bCompared to the broken-power-law model. ^c logpar model in XSPEC. ^dSlope at the pivot energy. ^eFixed pivot energy. ^fCurvature term.

Download table as:  ASCIITypeset image

2.2.2. Swift-X-Ray Telescope (XRT): X-Ray

The publicly available Swift -XRT data in the HEASARC database³² reveal that Swift observed 3C 279 51 times between 2013 November and 2014 April. We analyzed all those observation IDs (ObsIDs). The exposure times ranged from 265 s (ObsID:35019120) to 9470 s (ObsID:35019100). The XRT was used in photon counting mode, and no evidence of pile-up was found. The XRT data were first calibrated and cleaned with standard filtering criteria with the xrtpipeline software module distributed with the XRT Data Analysis Software (version 2.9.2). The calibration files available in the version 20140709 of the Swift-XRT CALDB were used in the data reduction. The source events were extracted from a circular region, 20 pixels (1 pixel $\simeq$ 236) in radius, centered on the source position. The background was determined using data extracted from a circular region, 40 pixels in radius, centered on (R.A., decl.: J2000) = (12^h56^m26^s, −05°49'30''), where no X-ray sources are found. Note that the background contamination is less than 1 % of source flux even in the faint X-ray states of the source. The data were rebinned to have at least 25 counts per bin, and the spectral fitting was performed using the energy range above 0.5 keV using XSPEC v.12.8.1. The Galactic column density was fixed at $2.2\times {10}^{20}$ cm⁻². The data were analyzed and the flux and photon index were derived separately for each ObsID. A relation between the unabsorbed model flux (0.5–5 keV) and photon index is represented in Figure 5. Only the ObsIDs with an exposure of more than 600 s and with more than 7 spectral points (= dof $\geqslant$ 5) were selected for the plot. A trend of a harder spectrum when the source is brighter is clearly detected.

Zoom In Zoom Out Reset image size

Data with ObsID of 80090001 and 35019132 were taken (quasi-) simultaneously with the NuSTAR observations, and here we report the details of those Swift-XRT observations. The observation with ObsID:80090001 was performed from 2013 December 17 21:06:51 to 22:39:56 (UTC) with an exposure time of 2125 section Therefore, this observation did not exactly overlap with the NuSTAR observation, but it was the closest available, starting about 17 hr after the end of the first NuSTAR observation of the source. The spectral fit with a PL model yielded a photon index of 1.67 ± 0.08 with a flux in the 0.5–5 keV band of $(12.2\pm 0.6)\times {10}^{-12}$ $\mathrm{erg}\ {\mathrm{cm}}^{-2}\ {{\rm{s}}}^{-1}$.

The other observation, with ObsID:35019132, was performed between 2014 January 01 00:20:26 and 22:50:54 (UTC), which overlaps well with the second NuSTAR observation. The exposure time of this Swift observation was 6131 s and the best-fit PL model displayed a photon index of 1.42 ± 0.03 with a flux in the 0.5–5 keV band of $(19.1\pm 0.3)\times {10}^{-12}$ $\mathrm{erg}\ {\mathrm{cm}}^{-2}\ {{\rm{s}}}^{-1}$. The spectral fitting results are reported in Table 3. Each snapshot observation during this ObsID was also analyzed separately. There were 15 snapshots in total and the exposure time in each snapshot was about 400 s typically, ranging from about 312 s to 594 s. The resultant count rates for all channels are shown in Figure 4.

2.2.3. Joint Spectral Fit of NuSTAR and Swift-XRT Data

Joint spectral fits of the NuSTAR and Swift-XRT data were performed for each NuSTAR observation. As described in previous sections, the data used for the spectral fitting were above 0.5 keV for the Swift-XRT and 3–70 keV for NuSTAR. Here, we introduced a normalization factor of order (1–3)% with respect to NuSTAR module A to account for differences in the absolute flux calibrations, and also to account for the offsets of the NuSTAR and Swift observing times (necessary for an analysis of a variable source). The Galactic column density was fixed at $2.2\times {10}^{20}$ cm⁻² as above. Simple PL,  BPL, and double-BPL (for the second observation) models were used for the source spectral models. The joint spectral fitting results are summarized in Table 3.

The joint spectrum during the first NuSTAR observation (December 16) can be represented by a simple power law from 0.5 to 70 keV with a photon index of 1.74 ± 0.01 (${\chi }^{2}$/dof = 688.6/679). The normalization factor of Swift-XRT with respect to NuSTAR module A is 1.01 ± 0.05. The BPL model improved the fit only marginally ($\sim 1\sigma$) with respect to the PL model. The result indicates that the X-ray spectrum of 3C 279 can be described by a single power law from the soft (0.5 keV) to the hard X-ray (70 keV) band, which is supported by the results from the individual fits for each Swift-XRT and NuSTAR observation alone.

For the joint spectrum during the second NuSTAR observation, the simple PL model did not result in an acceptable fit, with ${\chi }^{2}$/dof = 1072/996. Moreover, the normalization factor of Swift-XRT against NuSTAR module A was ∼0.7, which is clearly unacceptable. The BPL model, on the other hand, gave acceptable results with ${\chi }^{2}$/dof = 940.6/994, and the normalization of the Swift-XRT data was 0.97 ± 0.03. The break energy corresponded to 3.7 ± 0.2 keV with photon indices of 1.37 ± 0.03 and 1.76 ± 0.01, respectively, below and above the break energy.

This break energy is located where the bandpasses of Swift-XRT and NuSTAR data overlap, corresponding respectively to the higher and the lower energy end of those data sets. We are confident that the spectral break is a real feature for the following reasons. There is a significant difference in the photon index in each Swift-XRT and NuSTAR data set considered individually. This supports the conclusion that there is a spectral break at an energy close to the overlap of the Swift-XRT and NuSTAR bandpasses. Each resultant individual photon index is similar to the photon index derived from the joint fit below and above the break energy, respectively. The exposures of Swift-XRT and NuSTAR significantly overlapped (see Figure 4), yielding a reasonable inter-calibration constant (0.97 ± 0.03).

We also investigated a double-BPL model. The model yielded a probability of 0.22% ($\sim 3.1\sigma$) that the improvement in the fit was due to chance against the BPL model as assessed with an F-test. The second break energy appeared at ${9.4}_{-1.4}^{+2.1}$ keV and the photon index became even softer above the second break energy, changing from 1.72 ± 0.02 to 1.81 ± 0.02. The spectral break at that energy was also seen in the fitting result for the NuSTAR data only at ${9.3}_{-1.3}^{+1.6}$ keV. All these results suggest that the X-ray spectrum during the high state (the second NuSTAR observation) gradually softens with increasing energy, with the photon index changing by ∼0.4 from ∼0.5 to ∼70 keV. This is the first time that detailed, broadband spectral X-ray measurements of 3C 279 show spectral softening with increasing energy. Furthermore the absence of spectral softening in the first (December 16) NuSTAR observation clearly rules out the spectral shape as being caused by additional absorption. The joint X-ray spectral data points from the soft to the hard X-ray bands obtained by Swift-XRT and NuSTAR are plotted in Figure 6 in the $E\times F(E)$ (erg$\;{\mathrm{cm}}^{-2}\;{{\rm{s}}}^{-1}$) form. Finally, a log-parabola model was also tested using the logpar model in XSPEC. The pivot energy was fixed at 1 keV and the best-fit parameters are summarized in Table 3. The model also gave us acceptable fitting results, with ${\chi }^{2}$/dof = 960.6/995.

Zoom In Zoom Out Reset image size

2.3.1. Swift-UV/Optical Telescope (UVOT): UV Bands

The Swift-UVOT data used in this paper included all observations performed during the time interval from 2013 November to 2014 April. The UVOT telescope cycled through each of the six optical and ultraviolet filters ($W2$, $M2$, $W1$, U, B, V). The UVOT photometric system is described in Poole et al. (2008). Photometry was computed from a 5'' source region around 3C 279 using the publicly available UVOT ftools data-reduction suite. The background region was taken from an annulus with inner and outer radii of $27\buildrel{\prime\prime}\over{.} 5$ and $35\prime\prime$, respectively. Galactic extinction for each band in the direction of 3C 279 was adopted as given in Table 4.

Table 4.  Galactic Extinctions in the UV–Optical–Near-IR Bands as Used in This Paper

Band	${A}_{\lambda }$	Instruments
$W2$	0.271	UVOT
$M2$	0.285	UVOT
$W1$	0.195	UVOT
U	0.147	UVOT
B	0.123	UVOT, SMARTS
V	0.093	UVOT, SMARTS, Kanata
$R,{R}_{C}$	0.075	SMARTS, Kanata
IC,	0.056	Kanata
J	0.027	SMARTS
K	0.010	SMARTS

Notes. The extinctions are based on the reddening of $E(B-V)=0.029$ mag (Schlegel et al. 1998) with ${A}_{V}/E(B-V)=3.1$. See also Larionov et al. (2008)

Download table as:  ASCIITypeset image

2.3.2. SMARTS: Optical–Near-IR Bands

2.3.3. The Kanata Telescope: Optical Photopolarimetry

2.4.1. SMA: Millimeter-wave Band

The multiband light curves from the γ-ray to the radio bands taken between MJD 56615 and 56775, are shown in Figure 7 (covering the same period as in Figure 1). The γ-ray light curve measured by Fermi-LAT is plotted using 1 day time bins. The X-ray fluxes were measured by Swift-XRT in the 0.5–5 keV band. The third panel shows fluxes in the optical V-band measured by Swift-UVOT, SMARTS, and Kanata as well as the R-band data measured by SMARTS and Kanata. The optical polarization data were measured by Kanata in the R_C-band. The 230 GHz fluxes were based on the results from SMA and also included results by ALMA.³⁶ In the plot, the periods (A–D) as defined in Table 2 are also indicated.

Zoom In Zoom Out Reset image size

Generally, the source showed the most active states in the γ-ray band at the beginning (including Period B, Flare 1) and the end (including Period D, Flare 3) of the epoch considered in this paper. In the X-ray band, we also see two high-flux states, in the first half and in the second half of this epoch. While in the first active phase the flux variation was not apparently well correlated between the γ-ray and the X-ray bands, we can see flaring activities in both the γ-ray the X-ray bands around Period D (∼MJD 56750).

During the epoch considered here, the optical flux showed significantly different behavior than that in the γ-ray and the X-ray bands. In the beginning of this epoch, the measured fluxes were relatively low with relatively high polarization degrees, of $\sim 20\%$. Around period B, the γ-ray showed a very rapid flare with a hard photon index, but the source did not show any enhanced optical fluxes. After that, the optical fluxes started increasing gradually, with a drop of the polarization degree to $\sim 5\%$ after Period C. The γ-ray and X-ray band fluxes dropped, but the optical flux still continued increasing, and peaked at ~MJD 56720. In the largest flaring event in Period D, where the γ-ray ($\gt 100$ MeV) and X-ray fluxes were highest, the optical flux showed only minor enhancement, and had already started decreasing from its peak value. The optical polarization angle did not show any rotation throughout the observations considered here, and remained rather constant around $50^\circ$ with respect to the jet direction observed by Very Long Baseline Interferometry observations at radio bands (e.g., Jorstad et al. 2005).

The 230 GHz flux was less variable compared with other bands, varying by about 50%, from ∼8 to ∼12 Jy. Even though the amplitude of the variation was much smaller, the general variability pattern of the 230 GHz band followed a similar pattern to that seen in the optical; a low state in the beginning of the epoch, followed by increased activity in the middle, and a decrease toward to the end of the interval. No prominent millimeter-wave flares corresponding to the large γ-ray flaring events (Flares 1–3) were observed.

Figure 8 shows broadband SEDs for each period as defined in Table 2 (see also Figure 7 in the light curves). The data sets include Fermi-LAT (see also Figure 3), NuSTAR (for Periods A and C), Swift-XRT (for Periods A, C, and D), Swift-UVOT ($W1$ for Period A, $M2$ and V for Period C, all six bands for Period D), SMARTS (B, V, R, J, and K bands for all four periods) and Kanata (V, R_C bands for Period B), and SMA (for Periods A and C). All data in the figure were taken within the time spans as defined in Table 2, but with a very slight offset in some optical data as follows: for Period B (MJD 56646.4–56646.6), SMARTS data were taken during MJD 56646.348–56646.353 and Kanata data were taken starting at MJD 56646.8078. The SMARTS data observed during MJD 56750.1986–56750.2045 were for Period D (MJD 56750.210–56750.477). Unfortunately, no X-ray observation was performed during Period B. For comparison, the SEDs during a polarization change associated with a γ-ray flare observed in 2009 February, a low state in 2008 August (Hayashida et al. 2012), and very-high-energy γ-ray spectral points measured by MAGIC in 2006 (Albert et al. 2008) are also included.

Zoom In Zoom Out Reset image size

The γ-ray flares peaking at MJD 56646 (Flare 1, Period B) and MJD 56750 (Flare 3, Period D) are the brightest flares detected in 3C 279 by the Fermi-LAT. With photon fluxes of $\simeq 1.2\times {10}^{-5}$ $\mathrm{photons}\ {\mathrm{cm}}^{-2}\ {{\rm{s}}}^{-1}$ above $100\ \mathrm{MeV}$, they are brighter by a factor $\simeq 4$ than the flare peaking at MJD 54880 (Abdo et al. 2010a), and comparable to the record fluxes detected by EGRET (Wehrle et al. 1998).

Flare 1 was characterized by unprecedented rapid variability, with a flux-doubling time scale estimated conservatively at ${t}_{\mathrm{var}}\simeq 2\ {\rm{h}}$, and a very hard γ-ray spectrum, with photon index ${{\rm{\Gamma }}}_{\gamma }\simeq 1.7$. Notably, we do not detect any simultaneous activity in the optical band. These facts make Flare 1 different from any previous γ-ray activity observed in 3C 279. The closest analog was a rapid γ-ray flare in PKS 1510−089 peaking at MJD 55854 with a flux-doubling time scale of $\simeq 1\ {\rm{h}}$ and photon index of ${{\rm{\Gamma }}}_{\gamma }\simeq 2$ (Nalewajko 2013; Saito et al. 2013). However, in that case there were no simultaneous multiwavelength observations because of the proximity of PKS 1510−089 to the Sun.

The SED for Flare 1 is presented in Figures 8 and 9 as Period B. The very hard γ-ray spectrum measured by Fermi-LAT requires that the high-energy SED peak lies at energy $\gt 2\ \mathrm{GeV}$. We assume conservatively the actual SED peak lies at $\simeq 5\ \mathrm{GeV}$, so that the apparent γ-ray peak $\nu {L}_{\nu }$ luminosity is ${L}_{\gamma }\simeq 6\times {10}^{48}\ \mathrm{erg}\;{{\rm{s}}}^{-1}$. We further assume that this γ-ray emission is produced by the external radiation Comptonization (ERC) mechanism (see Sikora et al. 2009 for a review of alternative mechanisms). The corresponding synchrotron component is expected to peak close to the optical band. The observed simultaneous optical/UV spectrum, and the lack of simultaneous optical variability, place very strong constraints on the Compton dominance parameter $q={L}_{\gamma }/{L}_{\mathrm{syn}}\gtrsim 300$.

Zoom In Zoom Out Reset image size

In order to constrain the location along the jet r and the Lorentz factor ${{\rm{\Gamma }}}_{j}$ of the emitting region that produced Flare 1, we use the recent model of Nalewajko et al. (2014). In that model, the allowed parameter space for the γ-ray emitting region is defined by three constraints: (1) a jet collimation constraint ${{\rm{\Gamma }}}_{j}{\theta }_{j}\lt 1$, where ${\theta }_{j}$ is the jet half-opening angle, (2) a constraint on the SSC luminosity ${L}_{\mathrm{SSC}}\lt {L}_{{\rm{X}}}$, where ${L}_{{\rm{X}}}$ is the observed X-ray, hard X-ray, or soft γ-ray luminosity (depending on the expected energy of the SSC peak), and (3) a constraint on the radiative cooling time scale ${E}_{\mathrm{cool}}\lt 100\;\mathrm{MeV}$, where ${E}_{\mathrm{cool}}$ is the characteristic observed energy of the γ-ray photons produced by the electrons for which the radiative cooling time scale is comparable to the variability time scale. The proper size of the emission region is estimated directly from the observed variability time scale $R\simeq {\mathcal{D}}{{ct}}_{\mathrm{var},\mathrm{obs}}/(1+z)$, and it is related to the jet opening angle by $R/r={\theta }_{j}$. In addition to the parameters discussed above, we adopt a standard ratio of the Doppler-to-Lorentz factors ${\mathcal{D}}/{{\rm{\Gamma }}}_{j}=1$, and we also need to specify the upper limit on the expected synchrotron self-Compton (SSC) component. As the SSC component likely peaks in the MeV band, where we do not have any observational constraints, we will conservatively assume that ${L}_{\mathrm{SSC}}\lt {L}_{\gamma }/30\simeq 2\times {10}^{47}\ \mathrm{erg}\;{{\rm{s}}}^{-1}$. The constraints on the parameter space are shown in Figure 10. The yellow-shaded area indicates the allowed location of the γ-ray emitting region. For reasonable values of the jet Lorentz factor $20\lt {{\rm{\Gamma }}}_{j}\lt 30$, it should be located between $0.015\ \mathrm{pc}\lt r\lt 0.15\ \mathrm{pc}$, where the external radiation is dominated by the broad emission lines (${r}_{\mathrm{BLR}}\simeq 0.8\ \mathrm{pc}$). At larger distances, the energy density of external radiation fields will be insufficient to provide efficient cooling of the electrons producing the $100\ \mathrm{MeV}$ photons on the observed time scales, and also it will be more difficult to maintain a sufficiently high energy density to power such a luminous γ-ray flare from a very compact emission region. The Lorentz factor should be ${{\rm{\Gamma }}}_{j}\gt 17$, although this constraint would be stronger if we assumed a lower ${L}_{\mathrm{SSC}}$. For the subsequent modeling of the SED, we will adopt two possible solutions indicated in Figure 10: (B1) $r=0.03\ \mathrm{pc}$ and ${{\rm{\Gamma }}}_{j}=20$, and (B2) $r=0.12\ \mathrm{pc}$ and ${{\rm{\Gamma }}}_{j}=30$. This corresponds to the jet opening angle ${\theta }_{j}$ satisfying: (B1) ${{\rm{\Gamma }}}_{j}{\theta }_{j}=0.61$ and (B2) ${{\rm{\Gamma }}}_{j}{\theta }_{j}=0.34$.

Zoom In Zoom Out Reset image size

We use the Blazar code (Moderski & Sikora 2003) to model the SED of 3C 279 during Flare 1 with a standard leptonic model including the synchrotron, SSC, and ERC processes. The distribution of external radiation is scaled to the accretion-disk luminosity of ${L}_{d}\simeq 6\times {10}^{45}\ \mathrm{erg}\;{{\rm{s}}}^{-1}$ using standard relations for the characteristic radii of the broad-line region (BLR) and the dusty torus (Sikora et al. 2009) with covering factors ${\xi }_{\mathrm{BLR}}={\xi }_{\mathrm{IR}}=0.1$. In general, we use a double-BPL energy distribution of injected electrons $N(\gamma )\propto {\gamma }^{-{p}_{i}}$ with two breaks at ${\gamma }_{1}$ and ${\gamma }_{2}$ and three indices: p₁ for $\gamma \lt {\gamma }_{1}$, p₂ for ${\gamma }_{1}\lt \gamma \lt {\gamma }_{2}$, and p₃ for $\gamma \gt {\gamma }_{2}$. In order to reproduce a very hard γ-ray spectrum in the fast-cooling regime, we set ${p}_{1}=1$. The goal of the modeling is to match the ERC(BLR) peak with the observed γ-ray peak by adjusting the maximum electron Lorentz factor, ${\gamma }_{1}$, and the electron distribution normalization, and then to match the synchrotron component with the optical/UV data by adjusting the co-moving magnetic field B'. One should remember that because of the lack of any optical activity simultaneous with Flare 1, the synchrotron component should actually be below the optical/UV data points, so B' should be treated more like an upper limit than an actual value. The results of the SED modeling are presented in Figure 9, and essential model parameters are listed in Table 5. For solution B1 we obtain ${\gamma }_{1}=3700$ and $B\prime =0.31\ {\rm{G}}$, and for solution B2 we obtain ${\gamma }_{1}=2800$ and $B\prime =0.3\ {\rm{G}}$.

Table 5.  Parameters of the SED Models Presented in Figure 9

Model	A	B1	B2	C	D1	D2
$r\ (\mathrm{pc})$	1.1	0.03	0.12	1.1	0.03	1.1
${{\rm{\Gamma }}}_{j}$	8.5	20	30	10.5	25	30
${{\rm{\Gamma }}}_{j}{\theta }_{j}$	1	0.61	0.34	1	1	1
$B\prime \ ({\rm{G}})$	0.13	0.31	0.3	0.13	1.75	0.14
p1	1	1	1	1	1	1.6
${\gamma }_{1}$	1000	3700	2800	1000	200	100
p2	2.4	7	7	2.4	2.5	2.5
${\gamma }_{2}$	3000	⋯	⋯	3000	2000	6000
p3	3.5	⋯	⋯	3.5	5	4

Download table as:  ASCIITypeset image

We consider the basic energetic requirements for producing such an SED. We can estimate the total required jet power as ${L}_{j}\simeq {L}_{\gamma }/({\eta }_{j}{{\rm{\Gamma }}}_{j}^{2})$, where ${\eta }_{j}\sim 0.1$ is the radiative efficiency of the jet. And we can use the estimated magnetic field strength to calculate the magnetic jet power ${L}_{B}=\pi {R}^{2}{{\rm{\Gamma }}}_{j}^{2}{u}_{B}\prime c=(c/8){({{\rm{\Gamma }}}_{j}{\theta }_{j})}^{2}{(B\prime r)}^{2}$, where $R={\theta }_{j}r$ is the jet radius, ${\theta }_{j}$ is the jet half-opening angle, and ${u}_{B}^{\prime }=B{\prime }^{2}/(8\pi )$ is the magnetic energy density. For solution B1 we obtain ${L}_{j}\simeq 1.5\times {10}^{47}\ \mathrm{erg}\;{{\rm{s}}}^{-1}$ and ${L}_{B}\simeq 1.1\times {10}^{42}\ \mathrm{erg}\;{{\rm{s}}}^{-1}$, and for solution B2 we obtain ${L}_{j}\simeq 7\times {10}^{46}\ \mathrm{erg}\;{{\rm{s}}}^{-1}$ and ${L}_{B}\simeq 5\times {10}^{42}\ \mathrm{erg}\;{{\rm{s}}}^{-1}$. In both cases, the required magnetic jet power is a tiny fraction of the total jet power. This fraction is higher for solution B2, with ${L}_{B}{/L}_{j}\simeq 0.7\times {10}^{-4}$.This indicates that the emitting region responsible for Flare 1 is very strongly matter-dominated (cf. Janiak et al. 2015), although several observed SEDs of 3C 279 in 2008–2010 can be described by an equipartition model (Dermer et al. 2014).

The total required jet power ${L}_{j}$ appears very high compared with the accretion-disk luminosity ${L}_{d}$. For solution B1 we obtain ${L}_{j}/{L}_{d}\simeq 25$, and for solution B2 we obtain ${L}_{j}/{L}_{d}\simeq 11$. We note that there is no signature of increased disk luminosity in the UV data even for the lowest-flux state (Period A), which gives ${L}_{d}\lt 9\times {10}^{45}\ \mathrm{erg}\;{{\rm{s}}}^{-1}$. Assuming the black-hole mass of ${M}_{\mathrm{bh}}\simeq 5\times {10}^{8}{M}_{\odot }$ (see Section 1), the Eddington luminosity is ${L}_{\mathrm{Edd}}\simeq 8\times {10}^{46}\ \mathrm{erg}\;{{\rm{s}}}^{-1}$, hence ${L}_{d}/{L}_{\mathrm{Edd}}\simeq 0.08$ and ${L}_{j}/{L}_{\mathrm{Edd}}\simeq 0.9$ for solution B2. Solution B1 with moderate jet Lorentz factor ${{\rm{\Gamma }}}_{j}=20$ predicts a super-Eddington jet power. The predicted jet power can also be decreased when assuming a higher jet radiative efficiency ${\eta }_{j}\gt 0.1$. Taking this into account, the energetic requirements for this flare are consistent with the typical relation between ${L}_{j}$ and ${L}_{d}$ for FSRQs (Ghisellini et al. 2014).

In the standard scenario of magnetic acceleration of relativistic black-hole jets, the jets are initially dominated by the magnetic energy, which is gradually converted into the kinetic energy of plasma until the jet magnetization $\sigma \simeq {L}_{B}/({L}_{j}-{L}_{B})\sim 1$ (Begelman & Li 1994; Komissarov et al. 2007). Additional conversion of the jet magnetic energy is possible locally under special conditions, e.g., rarefaction acceleration induced by rapid decline in external pressure (Komissarov et al. 2010; Tchekhovskoy et al. 2010; Sapountzis & Vlahakis 2014). However, this additional jet acceleration is likely to lead to the loss of causal contact across the jet, and very wide opening angles (${{\rm{\Gamma }}}_{j}{\theta }_{j}\gt 1$), more typical of gamma-ray bursts. Magnetic energy can be dissipated directly in the process of magnetic reconnection. However, it is unlikely that relativistic reconnection can operate with sufficient efficiency to convert highly magnetized plasma with $\sigma \sim 1$ to very weakly magnetized plasma with $\sigma \sim {10}^{-4}$. Additional matter could be injected into the jet by individual massive stars crossing the jet (Khangulyan et al. 2013), or at the jet base during a brief pause in the jet production (Dexter et al. 2014). The latter possibility potentially allows for a more uniform distribution of an unmagnetized plasma layer across the jet.

The very hard electron energy distribution with ${p}_{1}\simeq 1$ extending up to $\gamma \gtrsim 2000$, required to explain the very hard γ-ray spectrum of Flare 1, is challenging for many particle acceleration mechanism and emission scenarios (e.g., Blandford & Levinson 1995). Very hard electron spectra can be obtained in relativistic magnetic reconnection, but they require extremely high electron magnetization ${\sigma }_{e}\gt 100$ (Guo et al. 2014; Sironi & Spitkovsky 2014; Werner et al. 2014). In the case of an electron–proton jet composition with no ${e}^{+}{e}^{-}$ pairs, one has ${\sigma }_{e}\simeq ({\bar{\gamma }}_{p}/{\bar{\gamma }}_{e})({m}_{p}/{m}_{e})\sigma$, where ${\bar{\gamma }}_{p}$ and ${\bar{\gamma }}_{e}$ are typical Lorentz factors of protons and electrons, respectively. In principle, it is possible that ${\sigma }_{e}\gg \sigma$, so that such extreme acceleration of electrons is possible even in the case of $\sigma \lesssim 1$. The final outcome of the acceleration depends on how the dissipated magnetic energy is shared between the protons and electrons; the first study of relativistic electron–ion reconnection suggests roughly equal energy division (Melzani et al. 2014).

We propose that Flare 1 of 3C 279, together with the similar flare of PKS 1510−089 peaking at MJD 55854 (Saito et al. 2013), constitute an emerging class of rapid γ-ray events characterized by flux-doubling time scales of a couple of hours, very hard γ-ray spectra with spectral peaks in the GeV band, and significant time asymmetry with longer decay time scales (Nalewajko 2013). Moreover, the results of this work indicate that such events do not have significant multiwavelength counterparts. Since only two clear examples were identified in bright blazars during ∼6 years so far of the Fermi mission, they appear to be rare events, and may not represent typical conditions of dissipation and particle acceleration in blazar jets.

In Figure 9, we also present two SED models for Flare 3 (Period D). This flare is characterized by a typical γ-ray spectrum, and a more typical Compton dominance, as compared to Flare 1. In addition, for Flare 3 we have simultaneous UV and X-ray data from Swift. The soft X-ray spectrum is very hard, with ${{\rm{\Gamma }}}_{{\rm{X}}}=1.22\pm 0.07$. We first attempted—in model D1 located in the BLR—to explain this X-ray spectrum by SSC emission from a very hard electron energy distribution (${p}_{1}=1$). By coincidence, model B1 described in the previous subsection does exactly that. However, since the γ-ray spectrum for Period D is much softer than the exceptionally hard γ-ray spectrum for Period B, in model D1 we need to adopt a break in the electron energy distribution at ${\gamma }_{\mathrm{br}}\lesssim 200$, which shifts the observed peak of the SSC component to $\sim 100\ \mathrm{keV}$; hence the X-ray part of the SSC spectrum is too soft to explain the observed X-ray spectrum. We note that Paliya et al. (2015) present an SED model for a period overlapping with our Period D, in which the X-ray spectrum is matched with the SSC component. They made the model by adopting a higher value of ${\gamma }_{\mathrm{br}}$, which requires a superposition of ERC(BLR) and ERC(IR) components to explain the γ-ray spectrum. In model D2 we were able to explain the X-ray spectrum with the low-energy tail of the ERC emission. The emission region in model D2 is located outside the BLR, and the entire high-energy component is strongly dominated by the ERC(IR) emission. We adopted a higher jet Lorentz factor ${{\rm{\Gamma }}}_{j}=30$ and the low-energy electron distribution index ${p}_{1}=1.6$ for $\gamma \lt 100$ (see Table 5). While model D2 matches the observed X-ray spectrum, because it is located at relatively large distance $r\simeq 1.1\ \mathrm{pc}$, it predicts a rather long observed variability time scale of ${t}_{\mathrm{var},\mathrm{obs}}\sim 2\;\mathrm{days}$.

Such extremely high-flux spectra as observed in Periods B and D could in some cases extend to even higher energies, and may possibly be detectable by ground-based Cherenkov telescopes (MAGIC, H.E.S.S., VERITAS, CTA). Despite its moderate redshift, 3C 279 was detected twice by MAGIC (Albert et al. 2008; Aleksić et al. 2011a) before the Fermi era. In this work, we have argued that the γ-rays originate at a radius $\sim 0.1\ \mathrm{pc}$, which is comparable with the estimated size of the broad emission line region ${r}_{\mathrm{BLR}}$ based upon reverberation mapping campaigns of other AGNs (e.g., Bentz et al. 2006; Kaspi et al. 2007). This radius is also roughly comparable to the minimum radius from which the highest energy photon observed during our campaign—${E}_{\mathrm{obs}}=26.1\ \mathrm{GeV}$ ($E\sim 40\ \mathrm{GeV}$ in the quasar rest frame) in Period A—can escape without pair production. The highest energy photons detected in Periods B and D were $10.4\ \mathrm{GeV}$ and $13.5\ \mathrm{GeV}$, respectively (see Figure 1). We could not distinguish whether the non-detection of $\gtrsim 15\ \mathrm{GeV}$ photons was due to the absorption by the BLR photons or just the statistical limitation of the short integration time for the spectra. Our emission models indicate a sharp drop in the source intrinsic spectral shape at $\gt 10\ \mathrm{GeV}$ energies due to adopting a very steep high-energy electron distribution index ${p}_{2}=7$. In addition, the γ-ray emission above $10\ \mathrm{GeV}$ produced in the ERC(BLR) process is suppressed due to the reduction of the scattering cross section in the Klein–Nishina regime.

The importance of γ-ray absorption in the pair-production process depends on the abundance of soft photons produced in the jet environment. One source of soft photons is the emission lines radiated by the broad emission line clouds. The optical depth depends on the geometrical shape of the BLR, and is significantly reduced for the flat geometries (Tavecchio & Ghisellini 2012; Stern & Poutanen 2014). Specifically, the results of Tavecchio & Ghisellini (2012) calculated for ${L}_{d}=5\times {10}^{45}\ \mathrm{erg}\;{{\rm{s}}}^{-1}$ and for an intermediate geometric case indicate that absorption from BLR photons is not significant for γ-ray photons with ${E}_{\mathrm{obs}}\lesssim 50\ \mathrm{GeV}$ emitted at $r\sim {r}_{\mathrm{BLR}}$, and those with ${E}_{\mathrm{obs}}\lesssim 20\ \mathrm{GeV}$ at $r\sim 0.1{r}_{\mathrm{BLR}}$. However, other models of the BLR can be expected to set a larger lower bound on the emission distance scale r. Another source of soft photons is Thomson scattering by the hot inter-cloud medium that is commonly invoked to confine the clouds by ram pressure. If we adopt the spectral component associated with the accretion disk by Pian et al. (1999), then the radius of the "γ-sphere" (for the $\sim 40\ \mathrm{GeV}$ photons in the source frame) is $\sim 0.2(\langle{\tau }_{T}\rangle/0.01)\ \mathrm{pc}$, where $\langle{\tau }_{T}\rangle$ is the mean Thomson depth (e.g., Blandford & Levinson 1995).

The difficulties faced by the leptonic models do not necessarily mean that they should be abandoned, unless there exists a better alternative. Hadronic models have been applied to 3C 279 (Böttcher et al. 2009; Petropoulou & Mastichiadis 2012; Diltz et al. 2015); however, they always require an extremely large jet power of order ${10}^{49}\ \mathrm{erg}\;{{\rm{s}}}^{-1}$, which is difficult to reconcile with observations of radio galaxies and theories of jet launching (Zdziarski & Boettcher 2015).

The X-ray and hard X-ray behavior of 3C 279 revealed by Swift-XRT and NuSTAR appears to be quite complex. This is especially interesting since the origin of this emission in FSRQ-type blazars is in general controversial and poorly understood (Sikora et al. 2013). Here, we focus on constraining the mechanism of X-ray and hard X-ray emission observed during the two NuSTAR pointings (Periods A and C).

The X-ray flux observed during the NuSTAR pointings is relatively high for 3C 279, higher by a factor of ∼5 than typical X-ray fluxes measured during the 2008–2010 campaign (Hayashida et al. 2012). In Period C, the X-ray flux is higher by a factor of ∼2 than in Period A, and it also shows some intraday variations. A comparison of the broadband SEDs for Periods A and C suggests a roughly linear relation between optical, X-ray, and γ-ray variations; however, this may be misleading, as the optical flux shows a long-term systematic increase that is not evident in the X-ray and γ-ray light curves. With typical photon index ${{\rm{\Gamma }}}_{{\rm{X}}}\simeq 1.7$, the observed X-ray spectrum is relatively hard, with a spectral break observed in Period C at $\simeq 3.5\ \mathrm{keV}$. The Swift-XRT data show a clear anti-correlation between the soft X-ray photon index and the soft X-ray flux (see Figure 5).

The X-ray emission of FSRQs is typically attributed either to the SSC emission of medium-energy electrons or to the ERC emission of low-energy electrons. Adopting a one-zone model in the SSC scenario for the X-ray band, we attempted to explain the observed broadband SEDs for Periods A and C together with the synchrotron emission for the optical band and the ERC component for the γ-ray band. Figure 9 shows the SEDs for Periods A and C with the emission models based on the parameters in Table 5. Actually, the observed high X-ray flux and hard X-ray spectra challenge the SSC scenario. In order to match the relatively high observed X-ray luminosity with that expected to be produced via SSC, as well as the corresponding synchrotron component and the ERC components with a condition of ${{\rm{\Gamma }}}_{j}{\theta }_{j}=1$, one needs to adopt a rather low jet Lorentz factor ${{\rm{\Gamma }}}_{j}\simeq 8.5$ for Period A and ${{\rm{\Gamma }}}_{j}\simeq 10.5$ for Period C. A low jet Lorentz factor requires a more powerful jet; in the case of Period C with γ-ray luminosity ${L}_{\gamma }\simeq 4\times {10}^{47}\ \mathrm{erg}\;{{\rm{s}}}^{-1}$ we estimate ${L}_{j}\simeq {L}_{\gamma }/({\eta }_{j}{{\rm{\Gamma }}}_{j}^{2})\simeq 4\times {10}^{46}{({\eta }_{j}/0.1)}^{-1}\ \mathrm{erg}\;{{\rm{s}}}^{-1}$. The jet Lorentz factor can be higher, and the required jet power lower, by allowing that ${{\rm{\Gamma }}}_{j}{\theta }_{j}\lt 1$. Typical X-ray SEDs from the SSC component in FSRQs are flat; it is possible to obtain a hard SSC spectrum by choosing an electron energy distribution peaking at ${\gamma }_{\mathrm{peak}}\gtrsim 300$. However, in order to reproduce apparent linear relations between optical, X-ray, and γ-ray fluxes from Periods A to C, it was necessary to adjust the value of ${{\rm{\Gamma }}}_{j}$ to compensate for the quadratic dependence of the SSC luminosity on the synchrotron luminosity ${L}_{\mathrm{SSC}}\propto {L}_{\mathrm{syn}}^{2}$. Therefore, while it is possible to make a one-zone model that fits the observed SEDs for both Periods A and C with X-rays produced by the SSC process and with reasonable jet power, such a model cannot naturally account for the observed flux variations over the multiwavelength bands. We note that lower Lorentz factors ${{\rm{\Gamma }}}_{j}\simeq 10$ required for modeling the SEDs for Periods A and C, together higher Lorentz factors ${{\rm{\Gamma }}}_{j}\simeq 30$ required for modeling the SEDs for Periods B and D, could indicate the existence of a spine-sheath jet structure (Ghisellini et al. 2005). In such case the γ-ray flares would be produced in the fast spine and the bulk of X-ray and hard X-ray emission in the slow sheath.

In the ERC scenario, on the other hand, the observed X-ray flux could be dominated by the low-energy tail of either the ERC(BLR) or ERC(IR) components, depending on the location of the emitting region. In one-zone models, this would require the X-ray spectra to be related to the γ-ray spectra by a single spectral component. This would explain the apparent linear relation between the X-ray and γ-ray data, although there are insufficient X-ray observations in the current campaign to probe the direct correlation between X-ray and γ-ray fluxes. (During the previous campaign on 3C 279, no significant correlation was detected between the X-ray and γ-ray fluxes; Hayashida et al. 2012). Judging from the simultaneous X-ray, hard X-ray and γ-ray SEDs, it could be possible to connect them by a single spectral component, especially for Period A. This would require a PL extension of the X-ray spectrum all the way to the low-energy end of the γ-ray spectrum. This is possible only for the ERC(IR) component where no cooling break is expected. However, because the low-energy ERC(IR) emission would be produced deep in the slow-cooling regime, very little flux variability would be expected on daily time scales. The ERC(BLR) component is very likely to feature a cooling break, and possibly an additional low-energy spectral break at $2.6{({{\rm{\Gamma }}}_{j}/20)}^{2}\ \mathrm{keV}$ produced by trans-relativistic electrons ($\gamma \sim 1$).

We conclude that there is no one-zone leptonic SED model that can satisfactorily explain the production of X-ray emission observed by NuSTAR in 3C 279. Various alternative mechanisms can be proposed where the observed X-ray emission is produced at a different location from the optical and/or γ-ray emission. Observation of a transient spectral break at $\simeq 3.5\ \mathrm{keV}$ in Period C may indicate a superposition of two spectral components in the X-ray band. Similar spectral breaks observed in high-redshift FSRQs were interpreted as due to very strong absorption (Fabian et al. 2001); however, in our case this interpretation is challenged by the lack of a break in Period A, only two weeks earlier. Alternative mechanisms for the X-ray emission include IC emission from the accretion-disk corona, ERC, or synchrotron emission from the jet acceleration region, and hadronic mechanisms (Böttcher et al. 2009). However, for most mechanisms it may be challenging to explain the relatively high X-ray luminosity ${L}_{{\rm{X}}}\simeq 2\times {10}^{46}\ \mathrm{erg}\;{{\rm{s}}}^{-1}$ observed during Period C, a factor ∼3 higher than ${L}_{d}$. This certainly excludes the accretion-disk coronal emission proposed tentatively for PKS 1510−089 in the low state (Nalewajko et al. 2012). The "jet base" scenario, where luminous weakly beamed X-ray emission is produced at very short distances from the supermassive black hole at which the jet is not yet fully accelerated, is motivated by recent observations of misaligned FSRQs (Bostrom et al. 2014), and will be investigated in detail elsewhere.

The observed optical flux shows a systematic increase by a factor of ∼4 over a period of ∼80 days (MJD 56645–56725). Superposed on this trend are weak flares that correspond very poorly to the strong γ-ray flares. No similar systematic trend is seen in the X-ray and γ-ray light curves. This is in contrast with the good overall correlation between optical and γ-ray fluxes in 2008–2010 (Hayashida et al. 2012). The lack of overall correlation between the optical light curve and the radio (mm-band), X-ray, and γ-ray light curves suggests that they are produced by different populations of electrons, and most likely at different locations. Moreover, the apparent linear relation between SEDs for Periods A and C is merely a coincidence. As the overall radiative output from 3C 279 is always dominated by the γ-ray emission, the systematic long-term increase in the optical flux could be due to a systematic increase of the magnetic field strength, or a systematic decrease of the external radiation energy density. The former option would be problematic if the jet magnetization remains constant, which would lead to an increase in the total jet power, and ultimately to an increase in the γ-ray luminosity, which is not observed.