Structural Analysis of Galaxies from Image Decompositions · Structural Analysis of Galaxies from Image Decompositions Concepts Dimitri Gadotti (ESO)

Structural Analysis of Galaxies fromImage Decompositions

Concepts

Dimitri Gadotti(ESO)

Initial Considerations

1. This topic in the context of the School is a very interesting complement

DAGAL School – Marseille 2014Dimitri Gadotti (ESO)

Athanassoula+ 13

Ho+ 11; CGS


2. The field is evolving very rapidly in the last few years. A useful source is the 2013 Deconstructing Galaxies conference:

http://www.eso.org/sci/meetings/2013/morph2013.html


GIM2D, GALFIT, BUDDA, SIGMA, MegaMorph, Galapagos, IMFIT, BDBar,GALFIDL, GASP2D


3. It is easy to press enter and start plotting the results. To do it right… it is damn difficult! Why? Because:

● There are a number of details one might miss, and they can ruin the results. It is still a technique just leaving its infancy.

● There is a lot of subjective input. In this sense, from a mathematical viewpoint, there’s no right answer, it always depends on one’s input and goals. To go from the math to the physical meaning is the challenging step.

● Models are often inaccurate!

● One is fitting a large number of parameters! There is significant degeneracy between them.


Outline

1. Motivation

2. Modeling structural components

3. Input

● what components to fit● fitting bars, nuclear point sources and disk breaks● how to guess well input parameters

4. More examples and applications

5. Uncertainties and degeneracy

6. The nasty effects of dust

7. Fitting edge-on galaxies

8. Working with simulations


MotivationWhere do we want to go and how do we get there

Ho+ 11; CGS



Gadot

ti 0

9



Gadot

ti 0

9



Ingredients:

• A galaxy image● With good measures of background and PSF

• Input parameters● Decide how many structural components to fit● Critical step: ellipse fits are very helpful to get good first guesses

• A code● That will create a model minimizing the reduced χ2:


Galaxy structural components are usually modeled each as a series of concentric ellipses at a fixed position angle and axial ratio b/a. (Remember: ellipticity is 1 – b/a.)

Apart from the geometrical properties, the light distribution of each component should be described too. This can vary for the different components depending on the code used. Here this corresponds to BUDDA (de Souza, Gadotti & dos Anjos 04; Gadotti 08).

Models


nuclear point source

diskbar

bulge

The disk surface brightness radial profile can be described either by a single exponential or by a broken/double exponential (Freeman 70; see also Erwin+ 08). Disks with larger scale lengths have a flatter profile.

Models


μ

r

The bulge and bar surface brightness radial profiles are described by a Sérsic function (Sérsic 63; see also Caon+ 93).

Models


The Sérsic function

Models


Graham & Driver 05

A high Sérsic index n means a steep inner profile and a shallow tail at low surface brightness.

There is relatively little difference in the profile shapes between profiles with high values of n (n>2). This has implications for (i) disk-like bulges(ii) measure uncertainties

• n=4: de Vaucouleurs 48

• n=1: exponential

• n=0.5: Gaussian


Models


Many ellipticals and classical bulges can be well fitted with a de Vaucouleurs (n=4) profile. However, these stellar systems show a range of Sérsic indices (e.g. Gadotti 09; Kormendy+ 09). Therefore, fixing n at any given value will lead to an overestimation (underestimation) of e.g. re if the true value of the bulge Sérsic index is less than (more than) the fixed value (Graham & Prieto 99).

Fixing n=4 to fit ellipticals and classical bulges is wrong!


Models


Kim+, in prep.

The light distribution of bars can be well described with a Sérsic function. It can also reproduce the observed dichotomy between bars with flat and exponential profiles (Elmegreen & Elmegreen 85; Kim+, in prep.).

Some times it is necessary to account for the presence of a nuclear point source, usually modeled with the PSF.

We will stop at a maximum of 4 components for the moment, but more complex models are possible.

Models


InputHow many components does the galaxy have? Which components are these?

A list of possible components include (and are not restricted to):

Each of these structural components has different (though in some cases similar) formation histories and physical properties. The photometric bulge can actually be several of these, even simultaneously. A disk-like bulge can be any of the components number 5 through 9 in the list above, or any combination of them.


1. disk (thin/thick)2. classical bulge3. bar+box/peanut+barlens4. spiral arms5. inner disk6. inner bar7. inner spiral arms8. lens(es)9. nuclear ring10. inner ring11. outer ring12. stellar halo13. nuclear point source


How does one decide which components to fit?

1. Inspect the galaxy image

2. Inspect isophotal contours● DS9/Analysis/Contours● IRAF/IMEXAMINE: contour plot, type ‘e’, edit parameters with

‘epar eimexam’, default parameters usually work fine

3. Inspect surface brightness profiles● 2D (IRAF/IMEXAMINE: radial profile, type ‘r’, edit parameters with

‘epar rimexam’)● Ellipse fits (IRAF/ELLIPSE)● Radial cuts (IRAF/PVECTOR)






1 8

2 0

2 2

2 4

2 6

µ (m

ag a

rcse

c-2) g a l a x y

t o t a l m o d e lb a rb u l g ed i s k

0 1 0 2 0 3 0r ( a r c s e c )

- 101

gala

xy -

mod

el

I C 4 8 6R

r

μ





InputDo I need to include a central point source?


A central point source can be:

1. A type 1 AGN (check NED)2. A nuclear stellar cluster

It alters significantly the bulge profile, in particular the bulge Sérsic index. As a rule, only add it if you are absolutely sure you need it. (This rule actually applies to all components!)



Gadotti 08



Gadotti 08

How can one be sure? Usually the point source creates a very steep inner profile, and, of course, results in a large Sérsic index.

InputFitting bars


Gadotti 08

If your galaxy has a bar, you MUST model it! (Laurikainen+ 05; Gadotti 08). Otherwise, the resulting bulge model can be very wrong.

InputFitting bars


Gadotti 08

If your galaxy has a bar, you MUST model it! (Laurikainen+ 05; Gadotti 08). Otherwise, the resulting bulge model can be very wrong.

InputFitting bars


Gadotti 08

2D fits are better than ellipse fits to measure bar axial ratios. Ellipse fits result in systematically less eccentric bars (by ~ 20%).

InputFitting bars


Gadotti 11

The distribution of bar ellipticities obtained from 2D fits for nearly 300 local barred galaxies peaks at ≈0.6, indeed about 20% higher than the peaks found in studies based on ellipse fits (e.g. Marinova & Jogee 07; Menéndez-Delmestre+ 07; Barazza+ 08; Marinova+ 09).

InputFitting bars


Gadotti 08

Residual images show dust lanes reaching very close to the center.

InputFitting bars


Gadotti 08

Residual images also show nuclear spiral arms, excesses of light at the bar ends, and… barlenses! (See talks by Athanassoula and by Laurikainen at the Deconstructing Galaxies meeting.)

InputFitting bars


Gadotti 08

InputFitting bars


Gadotti 08

Elongated structure inside the bar may be pointing out a region dominated by more eccentric orbits. And maybe we need more sophisticated bar models…

Ath

an

ass

oula

92

InputFitting bars


Gadotti 08

Gadotti 08: “…residual images show another structure within the bar, but this is only in the central region of the galaxy. This could be associated with an inner disc or a lens.”

Lia and Eija suggest that barlenses are box/peanuts seen at close to face-on projections.

InputFitting bars


Gadotti 08

InputFitting bars


Gadotti 08

InputFitting disk breaks


Slide

by

Nach

o T

rujillo



Slide

by

Nach

o T

rujillo



Kim, Gadotti+ 14: structural analysis of 144 barred galaxies from S4G. Fits include disk breaks.



So, what happens if disk breaks are not accounted for?

Kim+ 14



So, what happens if disk breaks are not accounted for?

Erwin & Gadotti 12



Disk scale length and central surface brightness can be wrongly estimated.

Kim+ 14



Disk scale length and central surface brightness can be wrongly estimated.

Kim+ 14



This affects to some extent B/T, D/T and Bar/T.

Kim+ 14

InputAn example of a bad fit


Gadotti 08

InputHow to estimate good first guesses for all parameters?

• x0,y0: brightest pixel. E.g.,

● IRAF/IMEXAMINE: type ‘a’ near the center of the galaxy



15

20

25

µ (m

ag a

rcse

c-2)

n=4µ

e-µ0=8.3

re

8.3

mag

E

seeing!

n=3µ

e-µ0=6.2

re

h S0

15

20

25

µ (m

ag a

rcse

c-2)

n=2µ

e-µ0=4

re

h

4 m

ag

1 mag

early S

n=1µ

e-µ0=1.8

re

h late S

2 4 6 8r (kpc)

15

20

25

µ (m

ag a

rcse

c-2)

n=0.55µ

e-µ0=0.8

0.8 mag

re,Bar

LBar

early SB

bulgebardisktotal

2 4 6 8r (kpc)

n=1µ

e-µ0=1.8

re,Bar

LBar

late SB

PA and ε are easy: can get them directly from ellipse fits for all components

Disk: μ0 and h

Bulge: μe, re and n

Bar: μe, re, n and LBar



15

20

25

µ (m

ag a

rcse

c-2)

n=4µ

e-µ0=8.3

re

8.3

mag

E

seeing!

n=3µ

e-µ0=6.2

re

h S0

15

20

25

µ (m

ag a

rcse

c-2)

n=2µ

e-µ0=4

re

h

4 m

ag

1 mag

early S

n=1µ

e-µ0=1.8

re

h late S

2 4 6 8r (kpc)

15

20

25

µ (m

ag a

rcse

c-2)

n=0.55µ

e-µ0=0.8

0.8 mag

re,Bar

LBar

early SB

bulgebardisktotal

2 4 6 8r (kpc)

n=1µ

e-µ0=1.8

re,Bar

LBar

late SB



15

20

25µ

(mag

arc

sec-2

)

n=4µ

e-µ0=8.3

re

8.3

mag

E

seeing!

n=3µ

e-µ0=6.2

re

h S0

15

20

25

µ (m

ag a

rcse

c-2)

n=2µ

e-µ0=4

re

h

4 m

ag

1 mag

early S

n=1µ

e-µ0=1.8

re

h late S

2 4 6 8r (kpc)

15

20

25

µ (m

ag a

rcse

c-2)

n=0.55µ

e-µ0=0.8

0.8 mag

re,Bar

LBar

early SB

bulgebardisktotal

2 4 6 8r (kpc)

n=1µ

e-µ0=1.8

re,Bar

LBar

late SB



15

20

25

µ (m

ag a

rcse

c-2)

n=4µ

e-µ0=8.3

re

8.3

mag

E

seeing!

n=3µ

e-µ0=6.2

re

h S0

15

20

25µ

(mag

arc

sec-2

)

n=2µ

e-µ0=4

re

h4

mag

1 mag

early S

n=1µ

e-µ0=1.8

re

h late S

2 4 6 8r (kpc)

15

20

25

µ (m

ag a

rcse

c-2)

n=0.55µ

e-µ0=0.8

0.8 mag

re,Bar

LBar

early SB

bulgebardisktotal

2 4 6 8r (kpc)

n=1µ

e-µ0=1.8

re,Bar

LBar

late SB


• Boxiness: the c coefficient here:

• Very important for bars

• Also for some early-type galaxies and bulges (e.g. Fornax A, ESO510-G13)



Athanassoula+ 90

c=2: perfect ellipse

c<2: disky ellipse a bar

c>2: boxy ellipse






Athanassoula+ 90

Kim

+,

in p

rep.






Athanassoula+ 90

Gadotti 11

Distribution of boxiness for nearly 300 local barred galaxies

Definitely important to fit c!



Athanassoula+ 90

Kim+, in prep.

Similar results found using 144 barred galaxies from S4G

• Central point source:

● Same FWHM as seeing.

● Peak intensity:

● Again can be measured with IRAF/IMEXAMINE, but need to subtract estimates for the central intensities of all other components.



• Image size in pixels• Pixel size

• Seeing FWHM + Moffat β parameter● Better than a Gaussian to account for atmospheric turbulence

(Trujillo+ 01)● 4.765 is canonical value● Very important to measure accurately! Affects significantly the

bulge Sérsic index.

• Background● Also, very important to measure accurately! Affects significantly

the disk scale length. Best to include background in the fitted image but not as a free parameter.

• Gain, read-out noise, number of combined frames


InputOther necessary input parameters


More Examples and ApplicationsGadotti 09


More Examples and ApplicationsGadotti 09


More Examples and Applications

Gadotti 09

Disk-like bulges, classical bulges and elliptical galaxies are clearly isolated in this diagram, indicating that the separation is not artificial, but has solid physical grounds.

A section with composite bulges can also be seen between classical and disk-like bulges.



Gadotti 09

How the mass-size relation of bulges and ellipticals compare?

log (size) = alpha × log (mass)

bars: α=0.21disks: α=0.33disk-like: α=0.20 (±0.02)classical: α=0.30ellipticals: α=0.38



Gadotti 09

● The mass-size relation of disk-like bulges is different from that of classical bulges by 5σ

● The mass-size relation of classical bulges is different from that of ellipticals by 4σ

● The only pair of components with similar mass-size relations are disk-like bulges and bars

bars: α=0.21disks: α=0.33pseudo: α=0.20 (±0.02)classical: α=0.30ellipticals: α=0.38

● At the high-mass end, classical bulges are not just ellipticals surrounded by disks

bars: α=0.21disks: α=0.33pseudo: α=0.20 (±0.02)classical: α=0.30ellipticals: α=0.38



Gadotti 09(See also papers by Laurikainen, Simard, and Graham.)



• ~ 3% in disk-like bulges

• ~ 4% in bars • ~ 32% in elliptical galaxies

• ~ 36% in disks

• ~ 25% in classical bulges

Gadotti 09

The stellar mass budget at redshift zero for galaxies with stellar mass > 1010 MSun

Uncertainties and Degeneracy

In BUDDA, 1σ uncertainties are calculated by varying each parameter individually until change in χ2 reaches the appropriate threshold. Uncertainties are usually around 10 – 20%. But this does note account for uncertainties in models and degeneracies (or coupling) between the various parameters and uncertainties.


Uncertainties and Degeneracy

Variations in input parameters can some times increase uncertainties. Bulge parameters (particularly n) are less stable.


The Nasty Effects of Dust

Dust is present in copious amounts in most disk galaxies, but also in a significant fraction of early-type galaxies. In the latter, it is usually interpreted as a sign of a merger event.

How does dust affect image decompositions?


ESO VISTA VVV Survey



NGC 2768 (SDSS)



NGC 2768 (HST, © Judy Schmidt)



Fornax A (“Taken from Bob and Pats farm in average seeing”, © Michael Sidonio). Ripples agree with merger scenario (Bosma+ 85).



Fornax A (HST)



BUDDA decomposition of a Ks SOFI NTT image reveals inner disk component (Beletsky, Gadotti+ 11).

Residuals after a bulge+AGN model only



Accounting for the disk reveals a nuclear, gaseous spiral-like structure, as well as a nuclear disk.

Residuals after a bulge+AGN+disk model

Beletsky+ 11



Modeling the gaseous spiral arms can reproduce nuclear undulations in geometric radial profiles.

Beletsky+ 11



VLT/SINFONI data reveal that the nuclear disk may be a kinematically decoupled core, and produces a sigma drop. Nowak+ 08 discuss the possibility of a distinct, younger stellar population there.

This is a very nice illustration of the power of image decomposition and residual images in unveiling hidden structural components.

Beletsky+ 11



A number of studies have addressed the effects of dust attenuation in the measurements of structural parameters of bulges and disks separately (e.g. Kylafis & Bahcall 87; Byun+ 94; Pierini+ 04; Möllenhoff+ 06; Driver+ 07).

These studies suggest that dust effects depend on:

1. Wavelength2. Galaxy inclination3. Star/dust geometry

In general, in the B band, dust can significantly affect:

•. disk scale length (resulting in an overestimation of up to 50%)•. central surface brightness, (dimming of up to 1.5 mag)

But what is the effect of dust in a decomposition, when both bulge and disk parameters are measured simultaneously?



In Gadotti+ 10 we have used SKIRT, a 3D Monte Carlo radiative transfer code (Baes+ 03, 05), to create artificial, dusty galaxies. Typical galaxy parameters were drawn from Hunt+ 04 and Gadotti 09.

Stellar population age defines spectral energy distribution.



Dust model is a mixture of graphite, silicate and polycyclic aromatic hydrocarbon dust grains. Physical processes include absorption, multiple anisotropic scattering and thermal re-emission. The V band face-on optical depth defines the total dust mass:

κV is the extinction coefficient at the center of the V band model. We considered models with τ = 0, 0.2, 0.5, 1, 2, 4, 6 and 8, and inclination angles of 0, 30, 45 and 80 degrees. And then we run BUDDA on them…



Gadotti+ 10



Gadotti+ 10



Gadotti+ 10



Gadotti+ 10



These results show that the disk luminosity distribution is strongly constrained by its outer parts, which contain a much larger number of pixels, where dust effects are relatively small. This produces an excess of light in the central regions of the galaxy, which is taken care of by making the bulge component less important. Because disk-like bulges are smaller, their galaxies suffer more strongly from dust effects.

Gadotti+ 10



In summary, there is a complex interplay between the bulge and disk models, the latter being less flexible to accommodate dust effects, resulting in:

1. Disk scale lengths are overestimated2. Bulge effective radii and Sérsic indices are underestimated

3. The attenuation in the integrated disk luminosity is underestimated, whereas the corresponding attenuation for the bulge is overestimated

4. All this results in B/D being underestimated

Bulge parameters are the most affected and the B/T ratio can be underestimated by a factor 2. Effects can be significant even at low inclinations and dust opacities. Further, effects can be stronger in galaxies with small, disk-like bulges. To derive accurate bulge parameters is difficult.

Edge-on Galaxies

One is dissecting ghosts… we can see through galaxies, there are projection effects, significantly strengthened in the case of edge-on galaxies.


NG

C 3

314 (H

ST)

Edge-on Galaxies

Dust effects are of course also enhanced.


NGC 891: 2MASS JHKs composite

Edge-on Galaxies

Identifying components is even less straightforward.


The COBE Project, DIRBE, NASA

Edge-on Galaxies



NGC 4251: SDSS gri composite

Edge-on Galaxies



NG

C 4

251:

SD

SS

i b

an

d (G

adot

ti+,

in p

rep.)

Edge-on Galaxies

Ellipse fits blend different components very strongly. Major/minor axes cuts might be more helpful.


Gadot

ti &

Sán

chez

-Jan

ssen

12

The Sombrero galaxy

Edge-on Galaxies

Disk surface brightness distribution has to be modeled differently, and scale height is an extra parameter to fit (van der Kruit & Searle 81).

One can guess z0 by cutting a profile perpendicular to the disk atR ~ h.


Edge-on Galaxies


Gadotti & Erwin, in prep.

Simulations

To compare results from real and simulated galaxies is a very instructive practice. E.g., one can see if simulations are reproducing observations. (Ongoing work with Lia and Taehyun Kim.)


Scannapieco, Gadotti+ 10

Cosmological hydrodynamical simulations (Aquarius; Springel+ 08) rendered with SUNRISE (Jonsson 06; Jonsson+ 10).

SUNRISE is a Monte Carlo radiative transfer code, producing images from the simulations, as if they were real observations: with dust, PSF, noise etc.

Simulations

The simulations aimed at reproducing Milky Way like galaxies.

• Haloes have no neighbor exceeding half of their mass within 1.4 Mpc

• Dark Matter particle: 2 × 106 M

• Gas particle: 3 × 105 M

• Same softening for all particles: 0.7 – 1.4 kpc

• Final halo masses: 7 – 16 × 1011 M

• Spin parameter: 0.008 – 0.049

• Gadget-3: stochastic star formation, metal-dependent cooling, chemical enrichment, feedback from Type II and Type Ia SNe, multiphase gas model (for details see Scannapieco+ 09)


Simulations

Some simulations show realistic bulges and bars, except for bulge effective radius.


Scannapieco+ 10

Simulations

D/T ratios measured photometrically alleviate disagreement between simulations and observations, as compared to kinematically measured D/T.


Scannapieco+ 10

Structural Analysis of Galaxies from Image Decompositions · Structural Analysis of Galaxies from Image Decompositions Concepts Dimitri Gadotti (ESO)

Documents