() - 1401.0538v1.pdf

arXiv:1401.0538v1 [astro-ph.HE] 2 Jan 2014

Mon. Not. R. Astron. Soc.

000

, 1–13 (2013)

Printed 6 January 2014

(MN L

X style file v2.2)

Connection between optical and

-ray variability in blazars

T. Hovatta

⋆

, V. Pavlidou

, O. G. King

, A. Mahabal

, B. Sesar

, R. Dancikova

S. G. Djorgovski

, A. Drake

, R. Laher

, D. Levitan

, W. Max-Moerbeck

, E. O.

Ofek

, T. J. Pearson

, T. A. Prince

, A. C. S. Readhead

, J. L. Richards

and J.

Surace

Cahill Center for Astronomy and Astrophysics, California I

nstitute of Technology, Pasadena CA, 91125, USA

Physics Department, University of Crete, PO Box 2208, 71003

Heraklion, Greece

IESL, Foundation for Research and Technology-Hellas, PO Bo

x 1527, 71110 Heraklion, Crete, Greece

Division of Physics, Mathematics, and Astronomy, Californ

ia Institute of Technology, Pasadena, CA 91125, USA.

Spitzer Science Center, MS 314-6, California Institute of T

echnology, Pasadena, CA 91125, USA.

National Radio Astronomy Observatory, P.O. Box 0, Socorro,

NM 87801, USA.

Benoziyo Center for Astrophysics, Faculty of Physics, Weiz

mann Institute of Science, Rehovot 76100, Israel.

Department of Physics, Purdue University, 525 Northwester

n Ave, West Lafayette, IN 47907, USA.

Accepted XXX. Received YYY; in original form ZZZ

ABSTRACT

We use optical data from the Palomar Transient Factory (PTF) and

the Catalina

Real-Time Transient Survey (CRTS) to study the variability of

-ray detected and

non-detected objects in a large population of active galactic nuclei

(AGN) selected

from the Candidate Gamma-Ray Blazar Survey and Fermi Gamma-Ra

y Space Tele-

scope catalogs. Our samples include 714 sources with PTF data and 1

244 sources

with CRTS data. We calculate the intrinsic modulation index to quantify

the opti-

cal variability amplitude in these samples. We find the

-ray detected objects to be

more variable than the non-detected ones. The flat spectrum rad

io quasars (FSRQs)

are more variable than the BL Lac objects in our sample, but the sign

ificance of the

difference depends on the sample used. When dividing the objects ba

sed on their

synchrotron peak frequency, we find the low synchrotron peake

d (LSP) objects to be

significantly more variable than the high synchrotron peaked (HSP)

ones, explaining

the difference between the FSRQs and BL Lacs. This could be due to t

he LSPs being

observed near their electron energy peak, while in the HSPs the emis

sion is caused by

lower energy electrons, which cool more slowly. We also find a significa

nt correlation

between the optical and

-ray fluxes that is stronger in the HSP BL Lacs than in

the FSRQs. The FSRQs in our sample are also more Compton dominated

than the

HSP BL Lacs. These findings are consistent with models where the

-ray emission of

HSP objects is produced by the synchrotron self-Compton mecha

nism, while the LSP

objects need an additional external Compton component that inc

reases the scatter in

the flux-flux correlation.

Key words:

galaxies: active – galaxies: jets – galaxies: BL Lacertae objects – g

alaxies:

quasars

1 INTRODUCTION

The extra-galactic

-ray sky is dominated by active galac-

tic nuclei (AGN). The spectral energy distribution (SED)

of AGN can be described by two components, a low en-

ergy component from radio to X-rays and a high energy

one from X-rays to very high energy

-rays. The low en-

⋆

E-mail: thovatta@caltech.edu

ergy component can be attributed to synchrotron radiation

in a relativistic jet while the high energy emission can be

either inverse Compton (IC) scattering of low-energy seed

photons by the synchrotron emitting electrons or emission

through a hadronic process. If the seed photons for the IC

scattering are the synchrotron photons, the process is call

synchrotron self-Compton (SSC) (e.g. Maraschi et al. 1992;

Bloom & Marscher 1996). Alternatively, the seed photons

can be external to the jet, for example, from the broad line

2013 RAS

T. Hovatta et al.

region or the molecular torus, in which case the process is

called external Compton (EC) (e.g. Dermer & Schlickeiser

1993; Sikora et al. 1994). For a recent study of the SED

modelling of AGN, see B ̈ottcher et al. (2013).

If an AGN is viewed with its jet very close to the line

of sight, it is called a blazar. Blazars emit brightly over th

entire electromagnetic spectrum, and can be further divide

based on their optical classification into flat spectrum radi

quasars (FSRQs) and BL Lac objects. The optical emis-

sion of BL Lac objects is dominated by a strong continuum

and they show only weak emission lines. FSRQs have strong

emission lines and are thought to have a much denser envi-

ronment near the black hole. The two classes also differ in

their large-scale jet properties. FSRQs are thought to be th

beamed counterparts of Fanaroff-Riley type II (FR II) galax-

ies, which have more powerful jets than the Fanaroff-Riley

type I (FR I) galaxies, thought to be the unbeamed coun-

terpart of BL Lac objects (Fanaroff & Riley 1974). However,

some recent studies have shown that the division of FSRQs

and BL Lacs into FR I and FR II galaxies does not always

follow this trend (e.g., Landt & Bignall 2008; Kharb et al.

2010).

Early studies using the Energetic Gamma-Ray Exper-

iment Telescope (EGRET)

-ray instrument showed that

high states in the optical and

rays are connected (e.g.,

Wagner et al. 1995a,b; Bloom et al. 1997; Hartman et al.

2001), but detailed comparisons were hindered by the poor

sampling of the light curves. The tendency for simulta-

neous flaring has been confirmed since the launch of the

Fermi Gamma-Ray Space Telescope (hereafter,

Fermi

) in

2008 and the capability of its Large Area Telescope (LAT)

(Atwood et al. 2009) to detect AGN even in their non-flaring

states. Many individual sources studied in great detail hav

shown correlated flares in optical and

-ray spectral regions

(e.g., Bonning et al. 2009; Marscher et al. 2010; Abdo et al.

2010a; Agudo et al. 2011; Ackermann et al. 2012b).

While individual sources have been studied in detail,

the number of studies including multiple sources is limited

Chatterjee et al. (2012) studied the 6 best sampled blazars

observed within the SMARTS blazar monitoring program

and found short delays between the optical and

-ray flares.

By decomposing the optical and

-ray light curves in to

individual flares they also found the shapes of the flares to

be similar in the two bands. However, Bonning et al. (2012)

studied 12 sources in the same program and found that in

some sources the variations were correlated while in others

they were not.

A statistical approach was taken by Arshakian et al.

(2012) who studied 80 LAT-detected objects in the MO-

JAVE sample. Using non-simultaneous optical data, they

found the optical and

-ray luminosities to be correlated in

the FSRQs in their sample, while no significant correlation

was found for BL Lac objects. The dispersion in the op-

tical to

-ray luminosity correlation was also much larger

than for radio and

-ray luminosities. They attributed this

to the possible larger variability in the optical compared t

radio, which would increase the scatter in non-simultaneou

correlations.

We use the data for 714 AGN in the Palomar Tran-

sient Factory (PTF) (Rau et al. 2009) and 1244 AGN in the

Catalina Real-Time Transient Survey (CRTS) (Drake et al.

2009) to study the connection between the optical and

ray emission in a

population

of AGN. The data sets have

637 sources in common. We use the intrinsic modulation

index (Richards et al. 2011) to study differences between

sub-populations of sources and the significance estimation

method of Pavlidou et al. (2012) to evaluate the flux – flux

correlation between the optical and

-ray data.

Our paper is organized as follows. In Sect. 2 we describe

our sample selection and data reduction. Sections 3 and 4

show the results of our analysis. We discuss our results in

the context of AGN models in Sect. 5 and summarize our

conclusions in Sect. 6

2 SAMPLE AND DATA REDUCTION

We use the AGN sample defined by (Richards et al. 2011)

and monitored by them at 15 GHz at the Owens Valley Ra-

dio Observatory (OVRO) 40-m as the starting point for our

analysis. This sample of 1771 objects includes 1158 sources

from the Candidate Gamma-ray Blazar Survey (CGRaBS)

(Healey et al. 2008) and all the sources above declination

−

◦

that have been detected by the LAT in the first and

second AGN catalogs (Abdo et al. 2010b; Ackermann et al.

2011). The OVRO sample also includes a small number of

objects with interesting jet properties or that are being mo

itored by other programs.

The CGRaBS sample is a statistically well-defined sam-

ple that was selected to resemble blazars that were detected

by EGRET. The sources were selected from a flat-spectrum

parent sample that is complete to 65 mJy flux density at

4.8 GHz and have radio spectral index

α >

−

5. The

CGRaBS sources were then selected based on their radio

spectral index, 8.4 GHz flux density, and X-ray flux from

ROSAT

All Sky Survey. The sample was compiled before the

launch of

Fermi

and was expected to contain a large num-

ber of sources that would be detected by

Fermi

. However,

Fermi

is much more sensitive to hard

-ray spectrum sources

than EGRET and a large number of CGRaBS sources have

not been detected by

Fermi

. This makes the sample ideal

for studying the differences between the

Fermi

-detected and

non-detected objects. Our PTF and CRTS samples contain

508 and 839 CGRaBS sources, respectively. We emphasize

that while our samples do not include all CGRaBS sources,

they are

unbiased

subsets of the complete sample as the

sample selection was not done based on the optical or

-ray

properties of the sources.

We search for all the objects within 1

′′

of our sample

targets in the PTF and 3

′′

in the CRTS. These limits were

selected based on the pixel size and typical seeing of the

observations. We describe the details of the data extractio

and analysis for each survey individually below. To allow

for comparison with simultaneous OVRO and

Fermi

obser-

vations, we use data from 2008 to 2013. The redshifts for the

sources in our sample range from 0.04 to 3.9 with a median

of 1. For the population studies we divide our final sam-

ples into

-ray loud and quiet objects based on their LAT

detection. We use only the sources in the CLEAN samples

of the 1st and 2nd LAT AGN catalogs (Abdo et al. 2010a;

here and throughout the paper we define the spectral index as

∝

2013 RAS, MNRAS

000

, 1–13

Optical and

-ray variability in blazars

Ackermann et al. 2011). These are sources for which only a

single association has been determined in the

Fermi

cata-

logs.

Additionally, we study the differences between the

FSRQ and BL Lac objects. The classifications we use come

mainly from the CGRaBS and

Fermi

catalogs. Furthermore,

we divide the sources based on the frequency of their syn-

chrotron peak

into low synchrotron peaked (LSP), in-

termediate synchrotron peaked (ISP) and high synchrotron

peak (HSP) objects. We use the classifications from the 2nd

LAT AGN catalog where the LSP sources have

Hz,

ISPs have 10

< ν

and HSPs have

(Ackermann et al. 2011). The number of sources in each sub-

sample is listed in Table 1.

2.1 PTF data

The Palomar Transient Factory is designed to observe op-

tical transients and variable sources. It uses the 48-inch

Samuel Oschin Telescope at Palomar Observatory with R-

and g

′

-band filters and a wide-field camera having a 7.2

square degree field of view. For a detailed description of the

project and its primary science goals see Law et al. (2009)

and Rau et al. (2009). We find PTF R-band data for 870 ob-

jects in our sample. The limiting magnitude is about 20.5.

Due to the nature of the PTF observations, some areas of

the sky get better coverage than others and therefore the

number of data points for each source depends on the sky

position and varies from less than 10 to a few hundred.

We use the PTF Photometric Pipeline to extract the

magnitudes. Images are processed using “standard” reduc-

tion procedures, including de-biasing, flat-fielding, and a

trometric calibration. Catalogs are generated using SEx-

tractor (Bertin & Arnouts 1996). The PTF data are photo-

metrically calibrated against the SDSS catalog. Photometr

nights are used in order to calibrate the photometry all over

the PTF footprints to accuracy of 2% (Ofek et al. 2012a,b).

In addition, we apply relative calibration to the photometr

with typical precision of a few millimagnitudes at the brigh

end (magnitude 15). The relative photometry algorithm is

described in Ofek et al. (2011).

We discard data based on various flags given by the

SExtractor (e.g., blending of multiple sources within the

field, bad astrometry of the field) and additional flags given

by the PTF Photometric Pipeline (e.g., saturated or dead

pixels). The sources can be observed up to four times each

night and we average the magnitudes over each night. This is

important because our variability analysis method assumes

subsequent observations to be independent and multiple ob-

servations within a night may introduce biases. We acknowl-

edge that AGN can be variable on time scales less than a

day (e.g., Wagner & Witzel 1995) but the variability ampli-

tude is typically a fraction of a magnitude. For our further

analysis we select all the objects that have at least three

data points. By visually examining all the light curves, we

also discarded sources which clearly suffered from blending

even if the data were not flagged by SExtractor. This results

in a sample of 714 sources which are listed in Table 5.

We correct the PTF R-band magnitudes for Galac-

tic extinction using the re-calibrated dust maps of

Schlafly & Finkbeiner (2011) with the reddening law of

Fitzpatrick (1999) extracted from NASA/IPAC Extragalac-

tic Database (NED). The corrected magnitudes were then

converted into flux density units (mJy) by using a zero point

of 3631 Jy. We note that there is an additional colour term

correction that affects the conversion (Ofek et al. 2012a) bu

it is less than 0.04 mag for a typical blazar spectrum and

therefore ignored. In all our further analysis we use the lig

curves in flux density units.

2.2 CRTS data

The Catalina Real-Time Transient Survey

uses the data

from the Catalina Sky Survey (CSS)

to report on opti-

cal transients. Observations are made without a specific fil-

ter with the 0.68-m Catalina Schmidt Telescope in Arizona,

USA, the 0.5-m Uppsala Schmidt at Siding Spring Obser-

vatory, NSW, Australia, and the 1.5-m reflector located on

Mt. Lemmon in Arizona. The main science goal of CSS is to

detect near-Earth objects but the large sky coverage makes

it ideal for transient studies. For details of the CRTS and th

first results see Drake et al. (2009), Mahabal et al. (2011),

and Djorgovski et al. (2012). We find CRTS data from the

Catalina Surveys Data Release 2 (Drake et al. 2009) for 1335

sources in our sample. The magnitudes for all the sources are

derived using the SExtractor program. The limiting magni-

tude of CRTS is about 21.

We discard data using the blend flag of SExtractor

which indicates if the source is blended with another one

in the field. Because the pipeline is not optimized for AGN

studies, there are many outliers in the data even after the

basic flagging. Outliers have a significant effect on our modu-

lation index analysis, so we discard the most extreme cases.

All the sources are observed four times each night within

∼

30 min time period. We use this additional information

to discard any data points which differ by more than 0.8

magnitudes from the other data points over the same night,

and all the data over a night if the standard deviation of the

data points exceeds 0.5 mag. These limits were empirically

determined from the data to exclude high outliers that are

unlikely to be due to intra-night variability. In the same wa

as for the PTF data, we average all the observations within

a single night. We then select all the sources with at least

three data points which results in a sample of 1244 objects.

All the sources are listed in Table 6.

For the modulation index study we must convert the

magnitudes into flux density units. In the case of the CRTS

data this is not straightforward because the observations a

done without a filter. However, the setup used resembles a

-band magnitude and it has been empirically determined

that the CRTS magnitudes can be converted into Cousins

-band by the relation

CSS

+ 0

(

−

)

+ 0

where (

−

) depends on the spectrum of the object. Be-

cause the (

−

) colour varies depending on the flux state

of the object (e.g., Bonning et al. 2012), and is unknown

for a large fraction of the sources in our sample, we take

a statistical approach in the conversion. Jester et al. (200

derive an empirical relation (

−

) = 0

(

−

) + 0

applicable for quasars at redshifts

2 using the SDSS data.

http://crts.caltech.edu

http://www.lpl.arizona.edu/css/

http://nesssi.cacr.caltech.edu/DataRelease/FAQ2.htm

2013 RAS, MNRAS

000

, 1–13

T. Hovatta et al.

Table 1.

Number of sources in the PTF and CRTS samples for the various s

ubsamples used in the analysis. There are 637 sources in

common between the two samples.

Sample

All

-ray loud

-ray quiet FSRQ BL Lac LSP ISP HSP

PTF

714

313

401

448

168 127

CRTS 1244

543

701

717

293 193

67 131

We extract (

−

) colours for 10 000 quasars using the SDSS

DR5 sky server SQL tool

to obtain a mean value of 0.34

for the (

−

) of quasars. We use this value to convert

the CSS magnitudes into Cousins

-band magnitudes, and

a zero point of 3953 Jy to convert into mJy. There are 123

sources in our CRTS sample with redshift larger than 2,

which are outside the nominal redshift range for the conver-

sion. We note that finding the exact conversion factor is not

critical because the multiplicative factor will cancel out

the calculation of the modulation index. However, this will

have an effect on the flux-flux correlation and therefore we

only use the PTF data in the correlation analysis.

3 INTRINSIC MODULATION INDEX

Variability amplitudes in light curves can be studied in

numerous ways, for example using the variability index

(e.g., Aller et al. 1992), the fractional variability ampli

tude

(e.g., Edelson et al. 2002), and the modulation index (e.g.,

Kraus et al. 2003). While all these methods are widely used,

they do not account for the effects of irregular sampling and

are often applicable only when the variations significantly

exceed the measurement errors. Furthermore, they do not

provide an estimate of the error in the variability measure.

The intrinsic modulation index

was first defined and

used in Richards et al. (2011) on the OVRO 40-m 15 GHz

data. Its basic principle is the same as the standard mod-

ulation index, defined as the standard deviation of the flux

density measurements in units of the mean flux density. The

main difference is that it is calculated using the intrinsic

standard deviation

and the intrinsic mean flux density

of the observation so that

(1)

The term intrinsic denotes values that we would obtain

if we had zero observational errors and infinite number of

samples. The intrinsic values are calculated using a likeli

hood approach which assumes the observed flux densities

to follow a normal distribution with Gaussian errors. The

measurement errors are accounted for in the calculation of

the joint likelihood for

and

. For full derivation of the

likelihoods see Richards et al. (2011).

The main advantage of this method is that it is pos-

sible to determine an uncertainty for the modulation index

which is needed in the comparison of the different source

populations. For sources with

within 3

of zero, the 3

upper limit will be determined. Another advantage is that

the number of data points in each light curve is accounted

for in the errors of

so that for sources with fewer data

http://cas.sdss.org/dr5/en/proj/advanced/quasars/qu

ery.asp

points the errors will be larger. The minimum number of

data points needed for the analysis is three. The intrinsic

modulation indices and the intrinsic flux densities are list

for each source in the PTF sample in Table 5 and in Table 6

for the CRTS sample.

There are some important limitations with the method

and the data that are illustrated in Fig. 1, which shows the

intrinsic modulation index plotted against the intrinsic m

ean

flux density of each object. One limitation is the systematic

errors in the data sets. It is often the case that sources that

are supposed to be constant in flux density will show non-

zero modulation index due to systematic uncertainties in

the data. In order to test this effect on the PTF and CRTS

data, we use data for white dwarfs, which are assumed to be

constant in the optical band, to determine the limit above

which the variations we detect are due to intrinsic changes

in the object and not due to systematic uncertainties. In

the PTF data, the highest intrinsic modulation index for a

white dwarf at a flux density above 0.3 mJy (see below for

the justification of this cutoff) is

= 0

035

002 and

therefore we set the lower limit of the intrinsic modulation

index to 0.04 for PTF blazars. Similarly, in the CRTS data

the highest value for a white dwarf at high fluxes is

078

02 and we set the lower limit to 0.08 for the CRTS

blazars.

For sources weaker than a certain flux density limit,

the method gives mainly upper limits and shows that below

the weak flux density limit, we cannot reliably determine

the intrinsic modulation index. This limit is 0.1 mJy for the

PTF and CRTS samples, estimated such that we include as

few marginal non-detections as possible (i.e. sources with

very high upper limits on the intrinsic modulation index).

This limit is also required because the white dwarfs show

a dependency on the flux density so that at lower fluxes

the limit for modulation index is higher, indicating that th

errors in the data are underestimated at low fluxes, making

the intrinsic modulation index estimates unreliable. Thes

limits are shown in Fig. 1 as yellow hatched regions.

There is also a large concentration of upper limits in

both plots at the lower left corner (cyan hatched region in

Fig. 1). This is a region where we cannot reliably determine

the intrinsic modulation index for sources that are fainter

than 0.3 mJy in the PTF sample and 0.5 mJy in the CRTS

sample. This effect disappears when the modulation index

is above 0.15 for the PTF sample and 0.25 for the CRTS

sample. In the following population studies we will exclude

the regions outside these limits (yellow and cyan hatched

regions in Fig. 1) to ensure that our results are not biased

due to incomplete sampling in these regions. We note that

our analysis is not sensitive to the exact value of the limits

and we have tested that even 50% higher flux density and

modulation index limits would not change any of our con-

clusions. After these cutoffs and excluding the upper limits

2013 RAS, MNRAS

000

, 1–13

Optical and

-ray variability in blazars

PTF mean flux density (mJy)

Intrinsic modulation index

0.01

0.1

100

1000

−

0.01

0.1

CRTS mean flux density (mJy)

Intrinsic modulation index

0.01

0.1

100

1000

−

0.01

0.1

Figure 1.

Intrinsic modulation index

with 1

uncertainty plotted against the intrinsic mean flux density

of all the sources in the

PTF (left) and CRTS (right) samples. Black triangles are 3

upper limits of

. Thick blue symbols are white dwarfs used as calibrators.

Yellow and cyan hatched areas show the regions not included i

n the population studies. See the text for details.

we have 271 sources in the PTF sample and 343 sources in

the CRTS sample.

Additional caveats for the method are listed in

Richards et al. (2011). One of the most important is the

assumption of Gaussian flux density distributions and the

leakage of probability density to negative flux densities wh

this is not met. It is clear that the flux density distribution

for many of our sources is non-Gaussian, especially when

there are only a few data points. This results in long tails in

the probability density distribution that in the case of hig

≥

5 leak to unphysical negative flux densities. We esti-

mate that for sources with

≥

7 about 8% of the “true

flux density” probability density leaks to negative values.

For sources with

≥

0 this becomes

∼

17%. There are 21

sources in the PTF sample and 73 in the CRTS sample with

m >

= 0

7. The solution to this problem is to extend the like-

lihood analysis to other shapes of flux density distribution

and will be presented in a forthcoming publication.

3.1 Redshift dependence

The redshift range for

-ray-detected sources is limited due

to the capability of

Fermi

to detect sources at high redshifts.

Similarly, BL Lac objects have typically lower redshifts th

FSRQs. In Fig. 2, we show

against the redshift for the

Fermi

-detected and non-detected FSRQs and BL Lacs in

both PTF and CRTS samples. The redshift range for

Fermi

detected sources extends to about

= 2 and the redshift

range of BL Lacs to

∼

1 while the non-detected sources

and FSRQs have a range of redshifts up to almost 4.

We have searched for correlations between

and red-

shift using the non-parametric Kendall’s tau test. FSRQs do

not show any significant correlations in the data. In BL Lacs

there is no significant correlation if we limit the redshifts

z <

1. Above this limit the redshifts of BL Lacs are incom-

plete (e.g. Shaw et al. 2013) and the correlation is biased. A

shown in Sect. 3.4 below, the mean intrinsic modulation in-

dices for FSRQs at

z <

1 and

z >

1 are similar to within 1

showing that there is no redshift dependence. These results

show that our conclusions are not affected by the different

redshift range of the populations, but for completeness, in

the following analysis of

-ray detected and non-detected

sources, and BL Lacs and FSRQs, we show the results for

both the total sample and samples restricted to a common

redshift range.

3.2 Mean intrinsic modulation index

In the following sections we will investigate whether the

intrinsic modulation index correlates with various physic

properties of the objects, such as the

-ray detection, opti-

cal classification, and location of the synchrotron peak. We

do this using a maximum likelihood approach which is de-

scribed in Richards et al. (2011). As shown in Fig. 3 the

distribution of

for the sources approximately follows an

exponential function. This allows us to calculate the prob-

ability density of the mean intrinsic modulation index

The errors of

are determined from the probability den-

sity distribution and they do not have to be symmetric. We

can then compare the means of the various sub-populations

and determine whether they agree within some significance

limit. We consider a result significant at the 3

level if the

p-value is less than 0.0027. The mean intrinsic modulation

indices for each sub-population are listed in Table 2.

In order to study the difference between the popula-

tions, we calculated the likelihood of the difference betwee

the mean intrinsic modulation indices in the two popula-

tions by taking the cross-correlation of the individual

likelihoods for each population. This allows us to estimate

2013 RAS, MNRAS

000

, 1–13

T. Hovatta et al.

0.0

0.5

1.0

1.5

Redshift

Intrinsic modulation index

−ray loud FSRQ

−ray quiet FSRQ

−ray loud BL Lacs

−ray quiet BL Lacs

PTF

0.0

0.5

1.0

1.5

Redshift

Intrinsic modulation index

−ray loud FSRQ

−ray quiet FSRQ

−ray loud BL Lacs

−ray quiet BL Lacs

CRTS

Figure 2.

Intrinsic modulation index against the redshift for the PTF

(left) and CRTS (right) samples.

Fermi

-detected FSRQs are

shown with black filled circles, non-detected FSRQs with ope

n black circles,

Fermi

-detected BL Lacs with filled orange triangles, and

non-detected BL Lacs with open orange triangles.

Table 2.

The mean intrinsic modulation indices and their 1

errors in the PTF and CRTS samples for the various sub-popula

tions used

in the analysis.

Sample

-ray loud

-ray quiet

FSRQ

BL Lac

LSP

ISP

HSP

PTF

201

015

−

014

096

015

−

012

215

023

−

020

149

015

−

014

277

035

−

030

199

038

−

031

096

017

−

014

CRTS 0

254

016

−

015

163

022

−

019

315

031

−

027

179

015

−

014

365

039

−

035

243

041

−

034

123

018

−

015

the most likely value for the difference between the popula-

tions. These are listed in Table 3.

3.3

-ray detected vs. non-detected sources

Figure 4 shows the probability density of the mean intrinsic

modulation index for the

-ray detected and non-detected

objects in the CRTS (top) and PTF (bottom) samples. The

two distributions are not consistent with a single value and

the

-ray detected objects are more variable than the non-

detected ones. Both PTF and CRTS give consistent results.

We find that the most likely difference between the mean

intrinsic modulation indices is more than 3

from zero.

If we limit the likelihood analysis to objects with red-

shift

z <

2 (the common range for the detected and non-

detected samples), the difference between the means of

the

Fermi

-detected (

PTF

= 0

230

023

−

020

and

CRTS

302

026

−

024

) and non-detected (

PTF

= 0

072

013

−

012

and

CRTS

= 0

161

027

−

022

) sources is still highly significant

with the most likely difference of

−

156

024

−

048

for PTF and

−

139

036

−

035

for CRTS.

One caveat is that the sample we use is not statisti-

cally complete as it is a combination of the CGRaBS and

Fermi

-detected sources which have different selection crite-

ria. Therefore we repeat the analysis using the CGRaBS

sources only, which is an unbiased sample. In this case the

difference between the

-ray detected and non-detected ob-

jects is even larger (see Table 3). This is mainly due to

Fermi

being more sensitive to HSP objects which are less variable

than the LSP sources that dominate the CGRaBS sample,

as we show below.

3.4 FSRQs vs BL Lac objects

Figure 5 shows the probability density of the mean intrin-

sic modulation index for the FSRQs and BL Lac objects in

the CRTS (top) and PTF (bottom) samples. Again, the two

distributions are not consistent with a single value and the

FSRQs are more variable than BL Lac objects. The most

likely difference between the mean intrinsic modulation in-

dex is more than 3

from zero for the CRTS sample and

nearly 3

from zero for the PTF sources. If we repeat the

analysis for the CGRaBS sources only, the most likely dif-

ference between the two populations is less than 1

(see

Table 3).

If we constrain our likelihood analysis to objects with

z <

1, we find that the means of FSRQs (

PTF

197

030

−

025

and

CRTS

= 0

318

041

−

035

) differ from the

BL Lacs (

PTF

= 0

152

026

−

021

and

CRTS

= 0

189

027

−

023

)

by less than 2

in the PTF with the most likely differ-

2013 RAS, MNRAS

000

, 1–13

Optical and

-ray variability in blazars

Table 3.

Most likely difference between the mean intrinsic modulatio

n indices and their 1

errors and significances for the various

sub-populations tested.

Populations

PTF Signif.

CRTS Signif.

-ray loud vs.

-ray quiet

−

103

020

−

020

−

090

027

−

025

CGRaBS

-ray loud vs. quiet

−

116

030

−

032

−

117

085

−

073

FSRQ vs. BL Lac

−

065

025

−

027

−

135

031

−

034

CGRaBS FSRQ vs. BL Lac

−

011

041

−

038

−

068

043

−

042

LSP vs. ISP

−

076

049

−

047

−

120

054

−

053

LSP vs. HSP

−

178

034

−

037

−

240

034

−

037

ISP vs HSP

−

101

035

−

041

−

119

038

−

044

intrinsic modulation index

probability density

0.0

0.5

1.0

1.5

PTF

0.0

0.5

1.0

1.5

0.0

0.2

0.4

0.6

0.8

1.0

intrinsic modulation index

cumulative distribution function

intrinsic modulation index

probability density

0.0

0.5

1.0

1.5

0.0

0.5

1.0

1.5

2.0

2.5

CRTS

0.0

0.5

1.0

1.5

0.0

0.2

0.4

0.6

0.8

1.0

intrinsic modulation index

cumulative distribution function

Figure 3.

Top: Histogram of the maximum-likelihood intrinsic

modulation indices

for the PTF sources with

m >

04 (left)

and for the CRTS sources with

m >

08, normalized as a proba-

bility density. The blue dashed line shows an exponential fu

nction

with a mean 0.18 (PTF) and 0.23 (CRTS), where the mean values

are determined using the maximum-likelihood analysis. Bot

tom:

Same data as in the top panel but shown with a cumulative dis-

tribution function.

ence of

−

044

036

−

037

and by about 2

in the CRTS with

the most likely difference of

−

127

044

−

047

. The lower sig-

nificance compared to the full sample is probably due to

the smaller number of FSRQs used in the limited analy-

sis. We test this by estimating the means of the FSRQs at

z <

1 and at

z >

1. The mean values at

z <

1 are listed

above and the most likely difference to the FSRQs at

z >

(

PTF

= 0

235

037

−

031

and

CRTS

= 0

309

050

−

041

) is less

than 1

for both the PTF sample with the most likely dif-

ference of

−

038

043

−

045

and the CRTS sample with the most

likely difference of

−

009

061

−

059

3.5 Division based on SED classification

Figure 6 shows the probability density of the mean intrinsic

modulation index for the LSP, ISP and HSP sources in the

CRTS (top) and PTF (bottom) samples. There is a clear

trend in both data sets for the LSP sources to be more vari-

CRTS

−ray loud

−ray quiet

0.0

0.1

0.2

0.3

0.4

PTF

−ray loud

−ray quiet

pdf

(

)

Figure 4.

Probability density of the mean intrinsic modulation

index for the

-ray detected and non-detected objects in the

CRTS (top) and PTF (bottom). Solid line is

-ray loud objects.

Dashed line is

-ray quiet objects.

FSRQ

BLLac

CRTS

FSRQ

BLLac

PTF

0.0

0.1

0.2

0.3

0.4

0.5

pdf

(

)

Figure 5.

Probability density of the mean intrinsic modulation

index for FSRQs and BL Lac objects in the CRTS (top) and PTF

(bottom). Solid line is FSRQs, and dashed line is BL Lac objec

ts.

2013 RAS, MNRAS

000

, 1–13

T. Hovatta et al.

LSP

ISP

HSP

CRTS

LSP

ISP

HSP

PTF

0.0

0.1

0.2

0.3

0.4

0.5

pdf

(

)

Figure 6.

Probability density of the mean intrinsic modulation

index for LSP, ISP and HSP objects in the CRTS (top) and

PTF (bottom). Solid line is LSPs, dotted line is ISP objects,

and

dashed line is HSP objects.

able than the ISP and HSP sources. In the case of LSP

vs. ISP sources we cannot distinguish the mean intrinsic

modulation indices from each other within 2

but the dif-

ference between LSP and HSP sources is highly significant.

The mean intrinsic modulation index difference is more than

from zero for both PTF and CRTS samples.

We also compare the intrinsic modulation index directly

to the peak frequency of the synchrotron component. We use

the peak frequencies used to determine the spectral classi-

fications tabulated in the 2LAC catalog (Ackermann et al.

2011, B. Lott personal comm.). In Fig. 7 we plot the intrinsic

modulation indices from the PTF sample against the peak

frequencies. The HSP sources are less variable than the LSP

sources that have a tail to much higher intrinsic modulation

indices.

4 CORRELATION BETWEEN OPTICAL AND

-RAY FLUXES

We use the method of Pavlidou et al. (2012) to study the

correlation between PTF and

-ray fluxes from the 2FGL

catalog (Nolan et al. 2012). The strength of the correlation

is quantified using the Pearson product-moment correlation

coefficient

while the significance of the correlation is de-

termined using simulated samples

. In order to account for

the common redshift in the two variables, the sampling is

done in the luminosity space. Because of this, we can only

use objects with known redshifts in the correlation analysi

Additionally, we only use simultaneous PTF and 2FGL data

and select those PTF sources that had at least 3 data points

between 2008-Aug-4 and 2010-Aug-1 (the integration period

The exact method for calculating the correlation coefficient

not critical because by estimating the significance through

simula-

tions, we ensure that the results do not depend on the distrib

ution

of the variables.

0.0

0.2

0.4

0.6

0.8

1.0

log synchrotron SED peak frequency [Hz]

intrinsic modulation index

FSRQ

BLLac

AGN

Figure 7.

Intrinsic modulation index from PTF plotted against

the peak frequency of the synchrotron component. FSRQs are

shown with open circles, BL Lac objects with filled orange cir

cles

and AGN without optical classification with black squares.

for 2FGL) to calculate the intrinsic mean flux density. This

results in a sample of 118 objects.

We first calculate the rest-frame optical luminosities for

all our objects using

(

) =

(

πd

(1 +

)

−

(2)

where

(

) is the PTF intrinsic mean flux density in mJy,

calculated using the likelihood method described in the pre

vious section,

is the luminosity distance to the object,

the redshift and

is the spectral index in the optical band

defined as

(

)

∝

. We do not have

values available

for the individual sources and use values

−

5 for LSP

objects and

−

1 for the HSP objects based on aver-

age values determined in Fiorucci et al. (2004). For sources

without SED classification or ISP objects we use

−

We note that the exact value of

does not have an effect

on the significance of the correlation since we use the same

value for both our data and the simulated samples.

The rest-frame

-ray luminosities at

= 1 GeV are

calculated using

(

) = (Γ

−

(

)

−

Γ+1

πd

(1 +

)

(3)

where

is the photon flux above

= 1 GeV and Γ is

the powerlaw index, both taken from the 2FGL. According

to Nolan et al. (2012) (see their Fig. 6) the photon fluxes

above

= 1 GeV are not affected by the spectral shape if

the photon flux is larger than 0

−

ph cm

−

. Our

sample includes only four sources below this limit (3 LSP

and 1 ISP) and therefore our sample is a nearly unbiased

subset of a flux limited

-ray sample.

We then construct simulated uncorrelated samples of

the same size as our true sample by pairing optical and

ray luminosities of different sources. We move back to the

flux-space by assigning a common redshift to the mixed pairs

and estimate the strength of the correlation in the intrinsi

cally uncorrelated sample using the Pearson correlation co

efficient. Whenever the sample size allows, we use multiple

2013 RAS, MNRAS

000

, 1–13

Optical and

-ray variability in blazars

Table 4.

Pearson product-moment correlation coefficient

for

the

-ray and optical flux correlations and the significance

the correlations in the various subsamples. Number of sourc

in each sample is also listed.

All

0.39

0.0043 118

FSRQ 0.46

0.0031 76

BL Lac 0.69 5

−

LSP

0.36

0.0229 69

ISP

0.94

0.0001 11

HSP

0.79

0.0005 19

redshift bins in the construction of the simulated samples.

In this way we are able to obtain uncorrelated samples that

are similar in dynamic range as our real observations. We

repeat this procedure 10

times to estimate the significance

(p-value) of the correlation. For a more detailed descripti

of the method and recommendations for the redshift bin-

ning, see Pavlidou et al. (2012).

Figure 8 shows the 2FGL

-ray photon flux against the

PTF mean optical flux density for the various subclasses.

The correlations, summarized in Table 4, are at least 3

sig-

nificant in all subclasses except for the LSP sources and all

sources together. There are some caveats in the interpreta-

tion of the correlations. Firstly we do not include upper lim

its in the optical or

-ray bands in the correlation estimation.

As shown in Lister et al. (2011), this can have a significant

impact on the result. We may be missing objects both in

the upper left (faint in optical) and lower right (

-ray faint)

corners of the plot which will affect the strength and the

scatter of the correlation, although not its significance. S

ec-

ondly, we have not accounted for the host-galaxy emission

in calculation of the optical fluxes, which may have an ef-

fect on the fluxes of the low redshift objects (e.g., Urry et al

2000; Falomo et al. 2003). Furthermore, the optical emissio

in FSRQs can have a significant thermal component (e.g.,

Raiteri et al. 2007) which may increase the scatter in any

correlations.

5 DISCUSSION

Using a large sample of sources from the PTF and CRTS

optical surveys, we have studied the variability propertie

-ray detected and non-detected objects, and of BL Lacs

and FSRQs, in addition to dividing the sources based on

their SED classification. We have determined the

intrinsic

modulation index

for the different source populations and

studied the flux - flux correlation between the optical and

rays.

There are 637 sources in common in the PTF and CRTS

samples. These include 320 sources with detected variabili

at more than 3

level. A notable result seen in Figs. 4-6

is that the mean intrinsic modulation indices of the CRTS

sources are systematically higher than in PTF. The number

of data points in the two surveys are comparable but the

CRTS data are more uniformly distributed over the five-year

period considered here. This indicates that in order to dete

the largest variability amplitudes long-term monitoring i

needed. However, all our main findings are seen in both PTF

0.00

0.05

0.10

0.15

0.20

0.25

0.30

100

pdf

(

)

CRTS optical

OVRO radio

Figure 9.

Probability density of the mean intrinsic modula-

tion index for OVRO and CRTS data. Solid line is CRTS data

(maximum-likelihood value and its 1

error

= 0

234

014

−

013

Dashed line is OVRO data (

= 0

125

004

−

004

and CRTS data sets, confirming the intrinsic nature of the

differences between the sub populations.

Our sample was selected from the monitoring sample

at OVRO so as to allow for comparison between the optical

and 15 GHz radio variability. We use the CRTS sample for

the comparison because it is larger (1244 objects) and the

more uniform sampling is more similar to the OVRO ob-

serving cadence (on average 2 observations per week). The

OVRO modulation indices are calculated based on four years

of data (Richards et al. 2014). Fig. 9 shows the probability

density of the mean modulation index for the overlapping

CRTS and OVRO samples. The variability in the optical

(

= 0

234

014

−

013

) is nearly twice as high as in the radio

(

= 0

125

004

). This is expected because the optical

emission is thought to originate in smaller emission region

and be caused by higher energy electrons near the peak of

the electron distribution.

5.1 Optical variability

One of our main results is that the

Fermi

-detected sources

are more variable in optical than the non-detected objects.

This has important implications for the identification of th

unassociated objects in the

Fermi

catalogs. Up to 30% of the

sources in the 2FGL catalog lack associations at other wave-

bands (Nolan et al. 2012). This is mainly because the source

position errors are still fairly large and the 95% error circ

can include multiple objects. In order to identify the cor-

rect counterpart for the

-ray source some additional infor-

mation is needed. Identifying unassociated sources has suc

cessfully been done by correlating the

-ray properties with

known source populations (Ackermann et al. 2012a) and by

using data at various other wavelengths (e.g., Kovalev 2009

;

D’Abrusco et al. 2013; Massaro et al. 2013a,b; Maselli et al.

2013).

Ruan et al. (2012) showed how the variability of sources

in optical catalogs can be used to identify the blazar coun-

2013 RAS, MNRAS

000

, 1–13

T. Hovatta et al.

−1.5

−1.0

−0.5

0.0

0.5

1.0

1.5

2.0

All

FSRQ

BLLac

−1.5

−1.0

−0.5

0.0

0.5

1.0

1.5

2.0

−2.0

−1.0

0.0

1.0

LSP

−2.0

−1.0

0.0

1.0

ISP

−2.0

−1.0

0.0

1.0

HSP

log PTF flux density [mJy]

log Gamma−ray photon flux [10

−

ph cm

−

]

Figure 8.

Integrated

-ray photon flux with its 1

uncertainty plotted against the intrinsic mean optical flux

density and its 1

uncertainty for all sources in our sample (top left), FSRQs (

top middle), BL Lac objects (top right), LSPs (bottom left),

ISPs (bottom

middle), and HSPs (bottom right). Sources without redshift

s are plotted in gray symbols and are not included in the calcu

lation of the

correlation.

terparts of the

Fermi

objects. They used optical data from

the LINEAR asteroid survey and characterized the variabil-

ity timescales and amplitudes of the objects. By selecting

sources with variability characteristics similar to those

known blazars, they were able to recover 88% of the known

associations in the 2FGL catalog, showing that identifying

blazars using optical variability is an efficient tool. Our re

sults agree with this conclusion and we suggest that the PTF

and CRTS data can be used to aid in the identification of

the unassociated sources (A. Mahabal et al. in prep.). One

advantage of our modulation index method is that it can be

used even for sources with only a few data points, as was

explained in Sect. 3.

Our result agrees with the analysis of the OVRO 15 GHz

radio data by Richards et al. (2011) that found the

-ray

detected objects to be more variable than the non-detected

ones. They suggest that one possibility is that all the radio

loud objects are also

-ray loud but were not detected by

the LAT because they were not flaring during the first year

Fermi

operations. This hypothesis was also suggested by

Kovalev et al. (2009) who showed that the sources detected

during the first three months of LAT monitoring were more

variable in the 15 GHz Very Long Baseline Array data than

the non-detected ones.

In addition, we divided our sources based on their op-

tical classification into BL Lacs and FSRQs and based on

their SEDs into LSP, ISP and HSP sources. When looking at

the total sample we find the FSRQs to be significantly more

variable than BL Lacs. This is mainly due to the location

of the SED peak so that the lower variability in BL Lacs is

driven by the less variable ISP and HSP sources. In fact, if

we compare the CGRaBS sources only (dominated by LSP

BL Lacs), or the

Fermi

-detected LSP BL Lacs and FS-

RQs, the difference between the two optical classes is sig-

nificant only at the 1

level. Similar results were obtained

by Ikejiri et al. (2011) who studied the NIR variability in

a sample of 42 objects. This could potentially be due to

many of the LSP BL Lacs being intrinsically similar to FS-

RQs as suggested by Giommi et al. (2012, 2013), but falsely

classified as BL Lacs due to the jet emission swamping any

thermal emission.

A similar tendency for higher variability in the LSP

sources was found in the

-ray data in the 2nd LAT AGN

catalog (Ackermann et al. 2011). They attribute this to

the location of the high-energy peak with respect to the

Fermi

band. LSP sources observed by

Fermi

are observed

at greater energies than the inverse Compton peak and

therefore are produced by higher energy electrons which

cool faster and vary more. HSP sources on the other hand

are observed at lower energies than the peak and therefore

would vary less. This is similar to what we see in the opti-

cal observations with respect to the synchrotron peak (e.g.

Mastichiadis & Kirk 1997; Kirk & Mastichiadis 1999). In-

terestingly, Richards et al. (2011) found using 15 GHz data

that BL Lac objects were more variable than FSRQs in

the CGRaBS sample. This can at least partially be ex-

plained by the CGRaBS BL Lac sample being dominated

by LSP objects which tend to vary more than the HSPs.

When a gamma-ray–selected parent sample is used instead

of CGRaBS, no significant difference in variability is found

between FSRQs and BL Lacs, and a trend with greater vari-

ability in LSPs than HSPs is seen (Richards et al. 2014).

Bauer et al. (2009) used the Palomar QUEST survey

to extract optical variability information for blazars and

also found that FSRQs exhibit larger amplitude variations

than the BL Lac objects. They suggest that this could be

2013 RAS, MNRAS

000

, 1–13

Optical and

-ray variability in blazars

due to higher jet power in the FSRQs which would allow

them to produce larger flares. This agrees with the sugges-

tion of Ghisellini et al. (1998) that more luminous objects

have higher energy density and therefore cool faster which

causes their synchrotron spectrum to peak at lower frequen-

cies. Ghisellini et al. (1998) used this to explain the so cal

led

blazar sequence (Fossati et al. 1998) where the peak of the

synchrotron and inverse Compton components shifts in fre-

quency and luminosity depending on the source type. Subse-

quently, several issues have been identified with the origin

blazar sequence (e.g., Padovani 2007; Nieppola et al. 2008)

and Ghisellini & Tavecchio (2008) updated their model to

include the accretion disc differences between the objects.

They suggest that the high power sources have an efficiently

radiating accretion disc that contributes to their emissio

while the lower power BL Lac objects have an inefficient

accretion disc. Recently Meyer et al. (2011) showed using

larger samples that, instead of a simple sequence the FS-

RQs and BL Lacs have different loci in the jet power - syn-

chrotron peak plane which can explain some of the differ-

ences in their observed properties. We may also be seeing

an additional contribution from the accretion disc in the

FSRQs in the optical band. Combining the effects of higher

jet power, more efficient accretion disc, and higher electron

energies at the optical band in FSRQs compared to BL Lacs

can most likely explain the difference in variability we see i

these objects.

5.2 Flux correlations

In addition to finding that the activity in the two bands

is connected, we also find the simultaneous fluxes to be

correlated. We find the correlation to have less scatter in

BL Lacs than the FSRQs. This is opposite to what was seen

by Arshakian et al. (2012) who used non-simultaneous data

to study the optical -

-ray flux correlation in 82 sources

and found a significant correlation only in the FSRQs. We

found that when testing for the correlation using the en-

tire PTF data set without requiring simultaneous data, the

scatter in the correlation was much larger. As suggested by

Arshakian et al. (2012) use of non-simultaneous data can

most likely explain the large scatter seen in their correla-

tions.

If the

-ray emission is produced by the synchrotron

self-Compton (SSC) method, we would expect a tight cor-

relation in the flux-flux space due to the same amount of

Doppler-boosting in these regimes. In the case of exter-

nal Compton emission, the high-energy component is more

boosted which destroys a linear dependence (e.g., Dermer

1995). Additionally, there can be a large contribution from

the external thermal photon field to the observed optical

fluxes, hindering any correlations further. It is typically

possible to model the SEDs of HSP sources using just a

single SSC component while the FSRQs and LSP sources

nearly always require an additional EC component (e.g.,

Lindfors et al. 2005; Abdo et al. 2010c, 2011; B ̈ottcher et al

2013).

Our observations are in good agreement with these

models with the ISP and HSP sources showing a tighter cor-

relation than the FSRQs. A Similar conclusion was drawn

by Lister et al. (2011) based on a tight correlation between

the

-ray to radio flux ratio and the synchrotron peak fre-

log synchrotron SED peak frequency [Hz]

Compton dominance

0.01

0.1

100

FSRQ

BLLac

Figure 10.

Compton dominance (ratio between

-ray and PTF

νF

) against the synchrotron peak frequency. Open circles are

FSRQs and orange filled circles BL Lac objects.

quency in a complete radio and

-ray selected sample of 173

AGNs. One caveat in the comparison using radio data is that

the radio and

-ray emission are produced by very different

energy electrons. The higher

-ray to radio ratios in HSP

objects in comparison to LSP sources seen in Lister et al.

(2011) and in Nieppola et al. (2011) can be explained by

the shifting of the synchrotron peak to higher frequencies i

HSP objects, which decreases the amount of radio emission

seen and increases the

-ray dominance.

Alternatively, the ratio between

-ray and optical emis-

sion is nearly the same as the Compton dominance, defined

as the ratio of the peak luminosity in the IC and synchrotron

components. As in Pavlidou et al. (2012), we use the photon

flux between 1 and 100 GeV tabulated in 2FGL to calculate

the

-ray flux density as

(

) = (Γ

−

(

)

−

Γ)

−

(4)

where Γ is the powerlaw index,

is the photon flux in units

ph cm

−

is the energy where we want to define the

energy flux density (in our case 1 GeV) and

is the lower

limit in energy (1 GeV in our case). Strictly, this equation

assumes the photon flux is integrated from

to infinity, but

due to the steep photon indices, the difference to limiting th

equation at 100 GeV is less than 1%. Expressing this in the

SED units in the rest frame of the source, we find

(

) =

(

)(1 +

)

(5)

where

is the energy at which to estimate the flux in erg

(in our case 1

−

erg corresponding to 1 GeV). Simi-

larly, we convert our simultaneous PTF mean intrinsic flux

densities to rest frame SED units at the central frequency

= 4

Hz (658 nm), which gives

νF

νS

(

)(1 +

)

−

)

(6)

where

(

) is the flux density in erg cm

−

, and

is the spectral index in the optical band. We then calculate

the ratio (

(

))

(

νF

(

)) which we call the Compton

dominance.

2013 RAS, MNRAS

000

, 1–13

T. Hovatta et al.

In Fig. 10 we plot the Compton dominance against

the synchrotron peak frequency. The LSPs are much more

Compton dominated than the HSPs as was also seen by

Abdo et al. (2010c). They attribute this to the higher EC

contribution in the LSP objects. This agrees very well with

our flux-flux correlation result. Our results are also very si

ilar to Finke (2013) who studied the Compton dominance in

Fermi

blazars by estimating the synchrotron and IC peak

SED luminosities. Finke (2013) explain the trends seen in

the plot with a simple model where the difference between

the sources is due to magnetic field and energy density in

the emission region, in addition to the viewing angle of the

source.

Taking the connected variations in the two bands to-

gether with the flux-flux correlation is a strong indica-

tion for SSC origin of

-ray emission in the HSP ob-

jects. However, we note that in individual sources and

in individual flares the situation may be more compli-

cated as shown in several detailed studies of individual

sources (e.g., Marscher et al. 2010; Ackermann et al. 2012b;

Nalewajko et al. 2012; Orienti et al. 2013).

6 CONCLUSIONS

We have studied the optical variability of AGN using large

samples of objects from the PTF and CRTS surveys. We use

a likelihood approach to calculate the intrinsic modulatio

index and a maximum likelihood method to study the dif-

ferences between various source populations. Additionall

we studied the flux-flux correlation between the optical and

-ray bands. Our main results can be summarized as follows:

(i) We find the

Fermi

-detected objects to be more vari-

able than non-detected ones. This shows that the activity

in the two bands is connected and that the optical variabil-

ity can be used as a tool for identifying unassociated

Fermi

sources.

(ii) The FSRQs in our total sample are more variable than

the BL Lac objects. This is likely due to the location of their

synchrotron peak because the mean modulation indices of

radio-selected CGRaBS FSRQs and BL Lacs or the

Fermi

detected FSRQs and LSP BL Lacs do not differ significantly.

(iii) When dividing the objects based on their syn-

chrotron peak location, we find the LSP objects to be more

variable than the HSPs, with ISPs in between the two. This

is similar to what is seen in the

-ray band and can be due

to differences in the electron energy in the observed bands.

(iv) We find a significant correlation between the optical

and

-ray fluxes which is tighter in the BL Lac objects.

The FSRQs are also more Compton dominated than the

BL Lacs. This is in accordance with models where the high-

energy emission of the HSP and ISP sources can be modeled

with a single SSC component while in FSRQs an additional

EC component is required.

ACKNOWLEDGMENTS

We thank the referee, Tigran Arshakian, for useful com-

ments that helped to improve the manuscript. T.H. thanks

Elina Lindfors for useful discussions. T.H. was supported i

part by the Jenny and Antti Wihuri foundation. V.P. is ac-

knowledging support by the European Comission Seventh

Framework Programme (FP7) through the Marie Curie Ca-

reer Integration Grant PCIG10-GA-2011-304001 “JetPop”

and the EU FP7 Grant PIRSES-GA-2012-31578 “EuroCal”.

The OVRO 40-m monitoring program is supported in part

by NASA grants NNX08AW31G and NNX11A043G, and

NSF grants AST-0808050 and AST-1109911. The National

Radio Astronomy Observatory is a facility of the National

Science Foundation operated under cooperative agreement

by Associated Universities Inc. The CSS survey is funded by

the National Aeronautics and Space Administration under

Grant No. NNG05GF22G issued through the Science Mis-

sion Directorate Near-Earth Objects Observations Program

The CRTS survey is supported by the U.S. National Science

Foundation under grants AST-0909182. This article is based

on observations obtained with the Samuel Oschin Telescope

as part of the Palomar Transient Factory project, a scien-

tific collaboration between the California Institute of Tec

nology, Columbia University, Las Cumbres Observatory, the

Lawrence Berkeley National Laboratory, the National En-

ergy Research Scientific Computing Center, the University

of Oxford, and the Weizmann Institute of Science. It is also

partially based on observations obtained as part of the Inte

mediate Palomar Transient Factory project, a scientific col

laboration among the California Institute of Technology, L

Alamos National Laboratory, the University of Wisconsin,

Millwakee, the Oskar Klein Center, the Weizmann Institute

of Science, the TANGO Program of the University System of

Taiwan, the Kavli Institute for the Physics and Mathematics

of the Universe, and the Inter-University Centre for Astron

omy and Astrophysics. This research has made use of the

NASA/IPAC Extragalactic Database (NED) which is oper-

ated by the Jet Propulsion Laboratory, California Institut

of Technology, under contract with the National Aeronautic

and Space Administration.

REFERENCES

Abdo A. A. et al., 2010a, Nature, 463, 919

Abdo A. A. et al., 2010b, ApJ, 715, 429

Abdo A. A. et al., 2010c, ApJ, 716, 30

Abdo A. A. et al., 2011, ApJ, 736, 131

Ackermann M. et al., 2011, ApJ, 743, 171

Ackermann M. et al., 2012a, ApJ, 753, 83

Ackermann M. et al., 2012b, ApJ, 751, 159

Agudo I. et al., 2011, ApJ, 726, L13

Aller M. F., Aller H. D., Hughes P. A., 1992, ApJ, 399, 16

Arshakian T. G., Le ́on-Tavares J., B ̈ottcher M., Torrealba

J.,

Chavushyan V. H., Lister M. L., Ros E., Zensus J. A., 2012,

A&A, 537, A32

Atwood W. B. et al., 2009, ApJ, 697, 1071

Bauer A., Baltay C., Coppi P., Ellman N., Jerke J., Rabinowit

D., Scalzo R., 2009, ApJ, 699, 1732

Bertin E., Arnouts S., 1996, A&AS, 117, 393

Bloom S. D., Marscher A. P., 1996, ApJ, 461, 657

Bloom S. D. et al., 1997, ApJ, 490, L145

Bonning E. et al., 2012, ApJ, 756, 13

Bonning E. W. et al., 2009, ApJ, 697, L81

B ̈ottcher M., Reimer A., Sweeney K., Prakash A., 2013, ApJ,

768, 54

Chatterjee R. et al., 2012, ApJ, 749, 191

2013 RAS, MNRAS

000

, 1–13

Optical and

-ray variability in blazars

D’Abrusco R., Massaro F., Paggi A., Masetti N., Tosti G., Gir

letti M., Smith H. A., 2013, ApJS, 206, 12

Dermer C. D., 1995, ApJ, 446, L63

Dermer C. D., Schlickeiser R., 1993, ApJ, 416, 458

Djorgovski S. G. et al., 2012, in T. Mihara, M. Serino, eds, Th

First Year of MAXI: Monitoring Variable X-ray Sources. p. 26

Drake A. J. et al., 2009, ApJ, 696, 870

Edelson R., Turner T. J., Pounds K., Vaughan S., Markowitz

A., Marshall H., Dobbie P., Warwick R., 2002, ApJ, 568, 610

Falomo R., Carangelo N., Treves A., 2003, MNRAS, 343, 505

Fanaroff B. L., Riley J. M., 1974, MNRAS, 167, 31P

Finke J. D., 2013, ApJ, 763, 134

Fiorucci M., Ciprini S., Tosti G., 2004, A&A, 419, 25

Fitzpatrick E. L., 1999, PASP, 111, 63

Fossati G., Maraschi L., Celotti A., Comastri A., Ghisellin

i G.,

1998, MNRAS, 299, 433

Ghisellini G., Tavecchio F., 2008, MNRAS, 387, 1669

Ghisellini G., Celotti A., Fossati G., Maraschi L., Comastr

i A.,

1998, MNRAS, 301, 451

Giommi P., Padovani P., Polenta G., Turriziani S., D’Elia V.

Piranomonte S., 2012, MNRAS, 420, 2899

Giommi P., Padovani P., Polenta G., 2013, MNRAS, 431, 1914

Hartman R. C. et al., 2001, ApJ, 558, 583

Healey S. E. et al., 2008, ApJS, 175, 97

Ikejiri Y. et al., 2011, PASJ, 63, 639

Jester S. et al., 2005, AJ, 130, 873

Kharb P., Lister M. L., Cooper N. J., 2010, ApJ, 710, 764

Kirk J. G., Mastichiadis A., 1999, Astroparticle Physics, 1

1, 45

Kovalev Y. Y., 2009, ApJ, 707, L56

Kovalev Y. Y. et al., 2009, ApJ, 696, L17

Kraus A. et al., 2003, A&A, 401, 161

Landoni M., Falomo R., Treves A., Sbarufatti B., Decarli R.,

Tavecchio F., Kotilainen J., 2012, A&A, 543, A116

Landt H., Bignall H. E., 2008, MNRAS, 391, 967

Law N. M. et al., 2009, PASP, 121, 1395

Lindfors E. J., Valtaoja E., M. T., 2005, A&A, 440, 845

Lister M. L. et al., 2011, ApJ, 742, 27

Mahabal A. A. et al., 2011, Bulletin of the Astronomical Soci

ety

of India, 39, 387

Maraschi L., Ghisellini G., Celotti A., 1992, ApJ, 397, L5

Marscher A. P. et al., 2010, ApJ, 710, L126

Maselli A. et al., 2013, ApJS, 206, 17

Massaro F., D’Abrusco R., Giroletti M., Paggi A., Masetti N.

Tosti G., Nori M., Funk S., 2013a, ApJS, 207, 4

Massaro F., D’Abrusco R., Paggi A., Masetti N., Giroletti M.

Tosti G., Smith H. A., Funk S., 2013b, ApJS, 206, 13

Mastichiadis A., Kirk J. G., 1997, A&A, 320, 19

Meyer E. T., Fossati G., Georganopoulos M., Lister M. L., 201

ApJ, 740, 98

Nalewajko K., Sikora M., Madejski G. M., Exter K., Szostek A.

Szczerba R., Kidger M. R., Lorente R., 2012, ApJ, 760, 69

Nieppola E., Valtaoja E., Tornikoski M., Hovatta T., Kotira

nta

M., 2008, A&A, 488, 867

Nieppola E., Tornikoski M., Valtaoja E., Le ́on-Tavares J.,

Hov-

atta T., L ̈ahteenm ̈aki A., Tammi J., 2011, A&A, 535, A69

Nolan P. L. et al., 2012, ApJS, 199, 31

Ofek E. O., Frail D. A., Breslauer B., Kulkarni S. R., Chandra

P., Gal-Yam A., Kasliwal M. M., Gehrels N., 2011, ApJ, 740,

Ofek E. O. et al., 2012a, PASP, 124, 62

Ofek E. O. et al., 2012b, PASP, 124, 854

Orienti M. et al., 2013, MNRAS, 428, 2418

Padovani P., 2007, Ap&SS, 309, 63

Pavlidou V. et al., 2012, ApJ, 751, 149

Raiteri C. M. et al., 2007, A&A, 473, 819

Rau A. et al., 2009, PASP, 121, 1334

Richards J. L., Hovatta T., Max-Moerbeck W., Pavlidou V.,

Pearson T. J., Readhead A. C. S., 2014, MNRAS in press

Richards J. L. et al., 2011, ApJS, 194, 29

Ruan J. J. et al., 2012, ApJ, 760, 51

Schlafly E. F., Finkbeiner D. P., 2011, ApJ, 737, 103

Shaw M. S. et al., 2013, ApJ, 764, 135

Sikora M., Begelman M. C., Rees M. J., 1994, ApJ, 421, 153

Urry C. M., Scarpa R., O’Dowd M., Falomo R., Pesce J. E.,

Treves A., 2000, ApJ, 532, 816

Wagner S. J., Witzel A., 1995, ARA&A, 33, 163

Wagner S. J. et al., 1995a, ApJ, 454, L97

Wagner S. J. et al., 1995b, A&A, 298, 688

2013 RAS, MNRAS

000

, 1–13