You are on page 1of 34

Grids of synthetic spectra for population synthesis

Paula R. T. Coelho
Institute of Astronomy, Geophysics and Atmospheric Sciences
University of São Paulo

pcoelho@usp.br
http://www.astro.iag.usp.br/~pcoelho
http://specmodels.iag.usp.br

Stellar Populations Meeting, Dec 15, São Paulo


Grids of synthetic spectra for population synthesis
Grids of synthetic spectra for population synthesis

Biased towards the work done at São Paulo, largely thanks to Beatriz legacy

Take a look at David Montes’ webpage for a complete list of libraries :):
http://pendientedemigracion.ucm.es/info/Astrof/invest/actividad/spectra.html
Our Grids of Synthetic Stellar Spectra
• Cayrel et al. ’91 (F, G and K stars)
• Barbuy ’94 (Mg2 for representative stellar evolutionary stages of globular clusters)
• Schiavon & Barbuy ’99 (λλ 6000-10200 Å, 2500<Teff<6000 K)

• SSPs: Milone et al. ’95 (TiO); Schiavon, Barbuy & Bruzual ’00 (NIR)
• besides work on line lists such as Melendez & Barbuy '00, Castilho et al. '99

• Barbuy et al. ’03 (λλ 460–560 nm, [α/Fe] = 0.0 and +0.4)
• Coelho et al. ’05 (λλ 300 — 1800nm, [α/Fe] = 0.0 and +0.4)
• Coelho et al. ’07 (first full spectra models for SSPs with α-enhancement)
• Coelho ’14 (λλ 300 — 900nm, 3000<Teff<25000 K,[α/Fe] = 0.0 and +0.4)

• Martins et al. ’05, ’07, ’14


Uses of libraries of stellar spectra

• deriving atmospheric parameters in stellar surveys, via automatic analysis and


classification of data

• determination of radial velocities via cross-correlation against templates, e.g. for the
detection of exoplanets

• calibration of features and indices for spectroscopic and photometric classification

• study of integrated light of galaxies as a core ingredient to stellar population models


stellar population
model stellar flux library

stellar evolution evolutionary


models synthesis code

SFH IMF
stellar population
model stellar flux library

it links what is
predicted with
what is observed

stellar evolution evolutionary


models synthesis code

SFH IMF
stellar population
model stellar flux library

it links what is
predicted with
what is observed

stellar evolution evolutionary


models Empirical synthesis code Theoretical
stellar flux library vs stellar flux library

Semi-empirical SP models Fully theoretical SP models

e.g. Leitherer et al. ’99, BC03


e.g. BC03 (high-res), Le Borgne
(03, low-res), González Delgado
et al. ’04 (PEGASE-HR),
SFH IMF
et al. 05, Coelho et al. ’07,
Vazdekis et al. ‘99, ’10
Percival et al. 09
stellar population
model stellar flux library

it links what is
predicted with
what is observed

stellar evolution evolutionary


models synthesis code

SFH IMF
stellar population
stellar flux library
model

radiative
transfer (line
evolutionary formation)
stellar evolution synthesis code code
models
stellar
atmosphere
model for a given
Teff, log g, etc

SFH
IMF
atomic and
molecular opacity
data
Empirical libraries Synthetic libraries

★The stars are real... ★ known atmospheric parameters


★ infinite S/N
★ large coverage in wavelength
★ high spectral resolution
★ S/N, flux calibration, telluric features
removal
★ compromise between wavelength
coverage and resolution
★ time consuming, demands
★ (in)accuracy of stellar parameters
computing power
(Teff, log g, Z)

★ coverage of the parameter space


in Teff, log g, metallicity and chemical ★ limited by the approximations and
pattern is limited (in)accuracies of the models
★ limited by the approximations and
Caveats of Synthetic Grids (in)accuracies of the models

• Physics:
• Sphericity, NLTE, 3D Hydrodynamics, etc…

• Opacities:
• The most sophisticated NLTE Hydrodynamical etc model will be wrong if the
opacities are wrong
• Missing lines, lines with wrong positions by several angstroms due to predicted
energy levels, molecular features which do not match observations…
Comparing models vs. observations

Spectral Indices
Fluxes
Atmospheric Parameters
Comparing models to observations

"Our first result is that it is not trivial to


compare model and empirical libraries because
errors in the atmospheric parameters of the
observed stars, and the particular abundance
pattern […] might mislead into wrong
conclusions about the accuracy of the models.
[…]
The library by Coelho employs a line list that
have been refined along the years in high-
resolution stellar spectroscopic studies, and the
effect is seen in its better average performance,
[…]
significant improvement can still be made by
fine tuning the atomic and molecular line lists,
through their calibration against high-
Martins & Coelho '07 resolution stellar spectra whose atmospheric
parameters – Teff , log g and detailed
abundance ratios – are known very accurately."
Tef f = 3800 ± 120 What are the wavelength regions
log g = 1.5 ± 0.3 Why not using them always then?
which systematically deviate?
Models Mean residual
Observations r.m.s.

Coelho '14

Tef f = 5250 ± 120


log g = 4.5 ± 0.1 (4000, 1.5)

Tef f = 11250 ± 400


log g = 4.0 ± 0.2
Residual Flux

(4750, 2.5)

(6500, 4.0)

Tef f = 21000 ± 400 (7000, 4.0)


log g = 4.0 ± 0.2

(11000, 4.0)

(11250, 4.0)

Wavelength Wavelength
too much opacity
missing opacity? (C2, CH, MgH)?

deeper core of H lines


(N-LTE, chromosphere)

For the modeller: where opacities


should be improved first?
For the user: which regions you
should mask/give lower weight?

Black lines: average empirical spectra minus model Below rms area: missing opacity
Orange areas: ± r.m.s. from empirical spectra Above rms area: too much opacity
4200 – 6800 Å Te↵ = 46 ± 231, log g = 0.15 ± 0.45, [Fe/H] = 0.03 ± 0.25
Safe fits
Not safe fits?

4828 – 5364 Å Te↵ = 73 ± 216, log g = 0.06 ± 0.45, [Fe/H] = 0.05 ± 0.20

Histograms of the Central values of the distributions are smaller


than typical uncertainties (no significant
residual (Model - Fitted) systematic differences)
Coelho '14
-1.45 ≤ [Fe/H] ≤ -1.15 -0.15 ≤ [Fe/H] ≤ +0.15

Safe fits
Not safe fits?

4828 – 5364 Å Te↵ = 73 ± 216, log g = 0.06 ± 0.45, [Fe/H] = 0.05 ± 0.20

Histograms of the Central values of the distributions are smaller


than typical uncertainties (no significant
residual (Model - Fitted) systematic differences)
Coelho '14
★ coverage of the parameter space
in Teff, log g, metallicity and chemical
Caveats of Empirical Grids pattern is limited

Coelho ’09
[α/Fe] and the need for
theoretical stellar

Peletier et al. '89


Worthey et al. '92
Figure from Worthey (1998)

McWilliam (1997)
[α/Fe] and the need for
theoretical stellar
Empirical Theoretical
vs
stellar flux library stellar flux library

Peletier et al. '89


Worthey et al. '92
Figure from Worthey (1998)

McWilliam (1997)
In other words…

We cannot use the star


formation history of a
galaxy like ...

to model the spectrum of a


galaxy like ...
McWilliam (1997)

We must rely on synthetic stellar spectra if we want to model the


spectra of galaxies different than our neighbourhood ...

… but we can do that in different ways.


α-enhancement in Stellar Population Models

Indices +
Fully theoretical models Differential models
Response Functions
✓Trager et al. ’00 ✓Coelho et al. ’07 ✓Walcher+ ’09

Thomas et al. (2003)

Vazdekis et al. 15
Schiavon(2007)

Percival et al. 09
Indices +
Fully theoretical models Differential models
Response Functions

✓e.g. Trager et al. ’00; ✓e.g. Coelho et al. ’07; Lee ✓e.g. Walcher+ ’09; Conroy
Thomas et al. ’03, ’11; et al. ’09; Percival et al. '09 +’12, 14; Vazdekis+'15
Schiavon ’07
✓Advantages: consistent ✓Most of the advantages of
✓Advantages: Widely treatment of evolutionary fully theoretical models
used in literature; quickly and spectral effects, full without their
model individual spectra and colours are disadvantages
element variations (C, N, modelled; all stars along
Ca, …) the isochrones are ✓Beware of: evolutionary
corrected effect not always included
✓ Beware of: corrections consistently; number of
are based in few stars, ✓Disadvantages: observables is more
evolutionary effects are uncertainties from model limited than fully
often not included stellar spectra; "expensive" theoretical; are the
consistently; only Lick to model individual corrections truly linear?
indices are modelled. abundance variations as it
needs a whole library of ✓ Becoming the gold
spectra and tracks standard nowadays
Sample of 2286 spectral
Early Type Galaxies from
SDSS-DR7

S/N > 40, no detectable


emission lines

r-band concentration
index larger or equal to
2.8

σ = 40 — 375 km/s

Spectral fitting performed


with code PARADISE
(Walcher et al. ’06) and
our differential models.

Archeology from integrated light Resolving the enrichment histories from


the integrated light of early type
(Walcher et al. 15) galaxies
Archeology in integrated
light, an example Walcher et al. (2015)

• Self-enriched early type galaxies:


their stellar populations are
consistent with internal chemical
evolution

• Accreted-mass early-type galaxies:


do not show a chemical evolution
signature, evidence of large
amounts of accreted material (be it
in form of gas or stars, merger or
infall)
Fully theoretical stellar population models
beyond α-enhancement

Anti-correlated abundances in globular clusters


Globular cluster ~
SSP? Not anymore

• GCs have been for decades seen as


Simple Stellar Population (SSP): all
stars having the same age and initial
chemical composition

• There is a growing body of


empirical findings that severely
challenges this traditional view (e.g.
Gratton et al. 12).

• But still used as benchmarks to test


SSP models!?

• The simplest depart from SSP are


chemical inhomogeneities, such as
anti-correlated variations of C-N, O-
Na, Mg-Al 47 Tuc
Photo credit: Thomas V. Davis
• Metal indices
appreciably
affected by the
abundance anti-
correlations are
Ca4227, G4300,
CN1, CN2 and
NaD

• Enhanced He
affects the Balmer
Observed 47Tuc line indices
(through the turnoff
Standard α-enhanced temperature),
CNONa variations mimicking a
population younger
by (up to) 2Gyrs.
Coelho et al. 09,11; Salaris et al. in prep.

Standard α-enh.: [Fe/H] = -0.7, [α/Fe] = +0.4


Exploring how anti-correlations
Modified CNONa: Typical of the upper end of
affect integrated spectra the observed pattern, [C/Fe] = -0.30; [N/Fe] =
+1.20; [O/Fe] = -0.45; [Na/Fe] = +0.60
on stellar spectroscopy
Name Age[Gyr] [Fe/H] Coelho et al. 11
[↵/Fe]
Walcher et al. 09
NGC6528 11±21 0.10 ± 0.22 +0.1±0.12
NGC6553 11±23 0.20 ± 0.14 +0.25 ±0.13
zing & Johnson (2002); 2 Zoccali et al. (2004); 3 Alves-Brito et al.
(2006); Meléndez et al. (2003)

PPLICATION TO GLOBULAR CLUSTERS


w test the newly prepared di↵erential models on high qual-
bular cluster spectra of NGC6528 and NGC6553 taken from
on et al. (2005) 2 . Only these two bulge GCs have [Fe/H] in
ge probed by the di↵erential models. For NGC6528 six spec-
available for consistency checks. Table 1 lists the published
nd abundances of these clusters. Not all literature measure-
have assigned errorbars and we have been forced to assign
C&N
lly realistic ones ad hoc. Mg
We fit the spectra of the GCs using the three di↵erent di↵er-
model sets (BC03+C07, V09+C07, PegHR+C07). While an
ive test of the C07 models is beyond the scope of this Let- Figure 2. The spectrum of NGC6528 (black), labelled a 2, as taken fro
also quote the results obtained with these models alone to Schiavon et al. (2005) and the best-fitting model from PegHR+C07 (red
From
te the Vazdekis
improvement et al. by
provided ’15theexercises
di↵erential models. The The blue line is the continuum that was used to normalize the observ
ting SSP is determined by simply computing 2 for every spectrum. The model was fit to the data only in the wavelength range 482
ith on
the spectral fitting:
three di↵erent parameters age, [Fe/H], [↵/Fe]. The 5364 Å.
urpose of using sedfit in this context is to automatically
or small shifts in velocity and for the resolution of the data.
n example fit is shown in Figure 2. Over the limited wave- the results for the other wavelength ranges scatter more and see
range of this fit, the model looks remarkable. A more in particular a↵ected by the age-[Fe/H] degeneracy. In the case
ative measure is the reduced 2 , which is between 2 (for R1, the available signal may not be strong enough. In the cases
528 b 1 with PegHR+C07) and 10 (for NGC6528 a 1 with large wavelength ranges, we expect that all models become less r
+C07). While thus the fit is not perfect, we note that an SSP liable, so small uncertainties in the modelling might dominate th
Conclusions #1
On synthetic spectra

• Comparison between synthetic and empirical stellar spectra (in the visible range):
flux differences below 2% for Teff >= 6000K and 10% at Teff = 4000K.

• Systematic differences exists at specific wavelength regions.

• Regarding automatic derivation of atmospheric parameters, synthetic and


empirical scales agree overall (see as well Dias et al. 2015, A&A 573, 13).
Conclusions #2
For population synthesis

• We do need synthetic stellar spectra in our galaxy models, if we want to


explore star formation histories different than our neighbourhood.
• Several “families" of spectral libraries are available in the market, and there is
not a “one-solution-fits-all”

Indices +
Fully Theoretical Models Differential models
Response Functions

• A model cannot be better than its ingredients! Take your time to choose and
understand your choice.
• We may need to rethink how to extract the chemical information from
integrated light? In the case of full spectrum fitting, beware of "competing"
chemical features and long baselines.
How does synthetic grids fit into MOS era?

• What do synthetic grids have to offer?


• Automatic Determination of Stellar Atmospheric Parameters
and Chemical Abundance

• What do synthetic grids have to gain?


• Large sample of high quality data to calibrate the models to

Thanks
Obrigada !

You might also like