Insights from Unsupervised Clustering Into the Physical Properties

Transcription

Insights from Unsupervised Clustering Into the Physical Properties
INSIGHTS FROM UNSUPERVISED CLUSTERING
ANALYSIS INTO QUASAR UV SPECTRA
Credit: ESO/M. Kornmesser
AYCHA TAMMOUR
UNIVERSITY OF WESTERN ONTARIO
GREAT LAKES QUASAR SYMPOSIUM
3 MAY 2016
SARAH GALLAGHER, GORDON RICHARDS, MARK DALEY,
KAREN LEIGHLY & NUR FILIZ AK
QUASAR WINDS
edge-on view
winds
~0.1c
X-ray source
Supermassive
black hole
accretion disk
~ 1016 cm
2
WIND SIGNATURES IN EMISSION LINES
STRONG
CIV
LOW SHIFT
WEAK
CIV
LOW SHIFT
WEAK
CIV
HIGH SHIFT
Richards et al. (2011)
3
X-ray vs. UV radiation
UV photons
drive the winds
high energy photons
over-ionize the gas &
suppress the winds
launching
radius
4
OPEN QUESTIONS
ARE THE WIND PROPERTIES UNIVERSAL AMONG ALL QUASARS,
AND OUR VIEW IS ONLY A MATTER OF THE LINE-OF-SIGHT
ORIENTATION TO THE WIND?
DO THE WIND PROPERTIES CHANGE WITH QUASAR PROPERTIES,
SUCH AS THE SHAPE OF THE IONIZING RADIATION?
5
QUASAR UV SPECTRUM
high ionization
intermediate
ionization
low ionization
6
LARGE DATASET
MULTI-DIMENSIONAL PARAMETER SPACE
PHYSICAL SPACE
7
K-MEANS
✦
✦
✦
Goal is to partition the parameter space into clusters of
similar objects
Create groups of equal variance, and minimize the withincluster sum-of-squares
Repeat until the centroids do not move significantly
8
QUASAR UV
EMISSION LINES
Paris et al. (2014) Quasar
Catalog :166,583 objects
equivalent width, redBHWHM
and blue-HWHM
Sample:
1.6 < z < 2.2
S/N >3
EW error < 10%
9
4,110
quasars
RHWHM
K-MEANS APPLIED TO C IV
Tammour A, Gallagher S. C., Daley M., Richards G. T.,
2016, MNRAS,459, 1659
10
GENERATE MEDIAN
COMPOSITE SPECTRA
FROM EACH CLUSTER
11
COMPOSITE SPECTRA FROM THE C IV CLUSTERS
12
K-MEANS APPLIED TO THE C III] BLEND
13
COMPOSITE SPECTRA FROM THE C III] CLUSTERS
14
DIVERSITY IN BROAD ABSORPTION TROUGHS
15
DIVERSITY IN BROAD ABSORPTION TROUGHS
SDSS BAL Quasar catalog
(Gibson et al., 2009):
5035 objects
measurements of BAL trough
strength,Vmin,Vmax
Our sample:
1.79 > z > 3.7
S/N >3
absorption in C IV
Vmax
Vmin
2,683 objects
16
K-MEANS APPLIED TO THE C IV TROUGH
CIV-f: low velocities and low equivalent width
CIV-a: high velocities and low equivalent width
Tammour et al. (in prep)
17
K-MEANS APPLIED TO THE C IV TROUGH
CIV-c and CIV-d: similar velocities but different
equivalent widths
Tammour et al. (in prep)
18
K-MEANS APPLIED TO C IV ABSORPTION TROUGH
Tammour et al. (in prep)
19
high velocities & low
equivalent width
high equivalent
width
PROPERTIES OF
CLUSTERS
low velocities &
low equivalent
width
20
PROPERTIES OF
CLUSTERS
SOFT
SED
HARD
SED
21
low fraction of lower ionization lines
~5% Si IV absorption
~0.5% Al III absorption
high fraction of absorption in
lower ionization lines
~50% Si IV absorption
~30% Al III absorption
PROPERTIES OF
CLUSTERS
low fraction of lower
ionization lines
~9% Si IV absorption
~1.5% Al III absorption
22
SOFT
SED & REDDENED
THICK WINDS
CIV-d
CIV-d
~ 1016 cm
23
SED
&
SOFT
HIGH VELOCITY OUTFLOW
CIV-a
CIV-a
~ 1016 cm
24
HARD
SED
WEAK WINDS
CIV-f
CIV-f
~ 1016 cm
25
QUESTIONS
CHOOSING NUMBER OF CLUSTERS
choose the elbow of cost function vs. number of clusters
EVALUATING
THE CLUSTERS
J=
N X
K
X
min(||xn
n=1 k=1
b a
s=
max(a, b)
1= dense, well-defined cluster
-1= less dense, overlapping
µk ||2 )
TESTING THE STABILITY OF OUR CLUSTERS
PROPERTIES OF CLUSTERS
Tammour et al. (in prep)
30
PROPERTIES OF CLUSTERS
Tammour et al. (in prep)
31