PyParadise Developed by: Bernd Husemann (MPIA), Omar Choudhury (AIP) � C. Jakob Walcher Leibniz Institut für Astrophysik Potsdam (AIP)
Physical properties • Physics 101: A physical property is described by – a number – a unit – an errorbar • Only well-defined quantities can be measured. • The “Star Formation History” is not a well-defined quantity. • The “contribution of stars aged between 1*10 9 and 5*10 9 yrs to the total luminosity in the V-band” IS a well-defined quantity. 24.11.2016 SELGIFS school, Walcher 2
Fitting Spectral Energy Distributions “Minimizing χ 2 is a maximum likelihood estimation of the fitted parameters if the measurement errors are independent and normally distributed. “ Press+, Numerical Recipes χ 2 is a measure of probability: P ( D | M ) ∝ e − χ 2 / 2 χ 2 compares a model and a dataset A model is a prior! There is no fitting without prior! 24.11.2016 SELGIFS school, Walcher 3
Fitting optical galaxy spectra: decompose into three Χ 2 n ◆ 2 ✓ F i − S i ∗ G ( A, v, σ kin ) χ 2 = X Kinematics of stars σ i i =0 A amplitude, v velocity, σ kin velocity dispersion ! 2 n F i − P M k =1 a k S i [ t k , Z k , T k ] Stellar populations χ 2 = X σ i i =0 a k weights, t k age, Z k Metallicity, T k Extinction ! 2 n F i − P L k =1 A k G i [ λ k , v, σ kin ] χ 2 = X Emission lines σ i i =0 A k amplitude, λ k line ID, v velocity, σ kin velocity dispersion Fi, Si, sigmai, n Number of pixels 24.11.2016 SELGIFS school, Walcher 4
Stellar populations considerations • Linear, but ill-posed problem – Regularization assumes something – Bootstrap your way through the problem • Transmission - Extinction issue – Normalize spectra (but restrict wavelength range) • Stellar populations accuracy ≠ continuum subtraction precision = kinematics determination – First case stelpop models – Second case stars better! 24.11.2016 SELGIFS school, Walcher 5
Pyparadise three main modules Non-linear (MCMC) fit of kinematics v, sigma, (h3, h4) Linear (NNLS) inversion for stellar populations <age>, <feh>, etc. (also per age bins) Non-linear (MCMC) fit of emission lines F , v, sigma 24.11.2016 SELGIFS school, Walcher 6
Pyparadise fiducial run Initial stelpop Kinematics fit Stelpop fit Kinematics fit Stelpop fit Emission line fit 7 24.11.2016 SELGIFS school, Walcher
Pyparadise bootstrap run Disturb spectrum, Do N subsample template basis times (Optional) Kinematics fit Stelpop fit (Optional) Emission fit 8 24.11.2016 SELGIFS school, Walcher
Pyparadise bootstrap run Disturb spectrum, Do N subsample template basis times (Optional) Kinematics fit Stelpop fit (Optional) Emission fit Mean, variance 8 24.11.2016 SELGIFS school, Walcher
Overall results • Stellar populations – Mean light weighted ages, Z, etc. for all templates – Mean light weighted ages, Z, etc. in bins of ages – Bootstrap-based errorbars – Weights of each SSP • Kinematics – Light weighted v, sigma – Bootstrap-based errorbars (incl. stelpop degeneracies) • Emission lines – Fluxes, v, sigma – Bootstrap-based errorbars (incl. stelpop and kin degeneracies) 24.11.2016 SELGIFS school, Walcher 9
Fit quality examples 24.11.2016 SELGIFS school, Walcher 10
Fit quality examples 24.11.2016 SELGIFS school, Walcher 11
Things to do when you fit spectra (and that you can do with PyParadise - but actually with any good software!)
Verify convergence! parameter value MCMC iteration number 24.11.2016 SELGIFS school, Walcher 13
Test your model! Our result Literature value Globular Cluster Spectrum Walcher+2009 Spectrum of NGC6553: Schiavon+05 See also Conroy and Vazdekis models 24.11.2016 SELGIFS school, Walcher 14
Test your method on the relevant parameter! Age [ α /Fe] Recovered [Fe/H] Z Input 24.11.2016 SELGIFS school, Walcher 15
Test applicability of Χ 2 statistics! theory data Walcher et al., 2015 24.11.2016 SELGIFS school, Walcher 16
Test applicability of Χ 2 statistics! theory data Walcher et al., 2008 24.11.2016 SELGIFS school, Walcher 17
Test applicability of Χ 2 statistics! theory data NO! Walcher et al., 2008 24.11.2016 SELGIFS school, Walcher 17
Look at your residuals! Rest-frame Observed- frame Walcher et al., 2015 24.11.2016 SELGIFS school, Walcher 18
Look at your residuals! 24.11.2016 SELGIFS school, Walcher 19
Look at your residuals! Template mismatch dominates! CALIFA Pipeline vs1.4 / DR2 24.11.2016 SELGIFS school, Walcher 19
Look at your residuals! Some Template molecular mismatch feature dominates! CALIFA Pipeline vs1.4 / DR2 24.11.2016 SELGIFS school, Walcher 19
There is science in template mismatch! 24.11.2016 SELGIFS school, Walcher 20
Do not trust good fits! observed spec fit around 4000A fit around 5000A 24.11.2016 SELGIFS school, Walcher 21
Do not trust good fits! observed spec fit around 4000A fit around 5000A 24.11.2016 SELGIFS school, Walcher 21
Do not over-interpret your results Do not trust details in your “star formation history”! Peaks are a consequence of Regularization is a possible solution, but imposes smooth SFHs, which may not be true 24.11.2016 SELGIFS school, Walcher 22
Do not over-interpret your results This MUST be wrong! Do not trust details in your “star formation history”! Peaks are a consequence of Regularization is a possible solution, but imposes smooth SFHs, which may not be true 24.11.2016 SELGIFS school, Walcher 22
Installing and running PyParadise
24.11.2016 SELGIFS school, Walcher 24
Recommend
More recommend