Two Examples from AGN Research about Flawed Data Analysis

1
Two Examples from AGN Research about Flawed Data
Analysis

• Gary Kilper
• Rice University
• AU Seminar
• 9/21/05

2
Overview of Talk

• Negative Evolution of HBLs
• Selection of HBLs in Sedentary Survey
• Analysis of number density of HBLs over different
  redshifts
• Variability analysis of AGN using SF
• Using the discrete-event model
• Results of SF analysis and comparisons with
  predictions

3
Negative Evolutionand ltV/Vmax gt

• Negative evolution higher number density or
  more luminous objects now than in the past
• ALL other AGN types exhibit positive evolution
  (see example above)
• Because of small samples, must use ltV/Vmax gt to
  measure evolution
• values lt ½ mean negative evolution

4
Typical BL Lac Spectra

• High-peaked
• HBL
• Blue
• Low-peaked
• LBL
• Red

5
Measured aox aro

6
Simulated aox aro for Surveys

• Radio selected X-ray selected

7
Object Selection in Sedentary Survey

• Restrictions result in a radio-flux-limited
  sample of extreme HBLs
• Radio-quiet sources are excluded

8
Data Cuts in Sedentary Survey

• 2074 unknown objects 1362 known
  AGN

9
Summary of Results for BL Lacs Surveys

10
Resultant ltVe/Vagt Measurements of Cosmological
Evolution

• Why would the radio flux limit affect the ltVe/Vagt
  for HBLs?

11
Shifting the HBL Spectrum at Higher Redshifts

• Shifts left by
• ?(z) ?0 (1z0) (1z)-1
• Shifts down by
• F(z) F0 dL(z0)2 dL(z)-2

12
Typical Approach to Finding ltV/Vmaxgt

• Cannot redshift the number of counts, which is
  what really limits the sample
• Instead must redshift the flux
• Flux is calculated using a spectral model
• The model assumes a power law tail (the dotted
  line) not an exponential cutoff
• (Notice the difference in figure on previous
  slide.)

13
The Necessary Approach to Calculating ltV/Vmaxgt
for HBLs

• Take a well-observed spectral source, with
  exponential cutoff at larger ?
• Redshift the spectrum and find Vmax where the
  simulated counts drop below detection limit of
  survey
• Compare to analysis using the Typical Approach
  on the same source
• Expand this to a sampling of objects

14
Part 2 Gauging AGN Variability A Well-Sampled
Light Curve

15
The Discrete Event Model

• Variability is due to superposition of
  independent events, occurring on random
  timescales
• Luminosity density L? kL N H? 2µ? C?
• Var(L?) kV N H?2 2µ?
• N event rate H? max amplitude
• 2µ? event duration C? non-var. lum.

16
The Structure Function and the Parameterized Fit

• Structure function of a light curve x(t)
• SFx(t) lt ( x(tt) x(t) )2 gt
• Fit to SF
• SFx(t) 2e2 A(t/tmax)a Btß (tlttmax)
• SFx(t) 2e2 A Btß (ttmax)
• tmax max. variability time scale (ß 2)
• e average uncertainty on light curves

17
Light Curves for Objects in Sample

• Simulated
• Light Curve

18
Resultant Structure Functions

19
Parameters of the Fits to the SFs

20
Problems with using the Structure Function for
this Sample

• Light curves have large gaps and there are not
  enough data points (unlike simulation)
• Here they even oversample by making the bin
  width larger than the bin spacing
• SF is dependent on sampling of data
• SF is measuring incompleteness of data
• Large error bars (which are not plotted) on SFs
• SFs tellingly show a plateau at twice the
  variance of the light curves

21
tmax over Different Wavelengths

22
Variability Time Scale and Event Rate over
Luminosity

• 2µ1300 ? L1300.21?.11 N ?/ L1300

23
Event Energy and Variability over Luminosity

• E ? L1300.99?.10 s(L) ? L1300.12?.11

24
Comparison to Physical Nature of AGN Variability
Events

• Black hole physical time scales
• Implies a 2µ ? L dependence (NO)
• Supernovae (starburst model)
• Should have no E L dependence (NO)
• Predicts s(L) ? L-1/2 (NO)
• Magnetic blobs above accretion disk
• Needs E ? L2 (NO)

25
The Stellar Collisions Model

• Proposed by Torricelli-Ciamponi, Foellmi,
  Courvoisier, Paltani in 1999
• Requires a relation of E ? L2/3
• However, here they claim that this model provides
  a relatively good approximation, although its
  more than a three sigma difference.
• Collision rate should go with MBH1/2, implying N
  ? L1/3 (NO)
• Yet they conclude that this model is favored.

26
Conclusions

• The unique (and therefore very important)
  negative evolution of HBLs still has not been
  thoroughly tested by Giommi Padovani
  preserving the extraordinary result even while
  knowing of the potential problems of selection
  bias and poor spectral modeling for six years.
• The research led by Paltani Courvoisier is
  biased towards proving the usefulness of their
  analytical technique, and the accuracy of their
  stellar collisions model.