The KIAA-Cambridge Joint Workshop on Near-Field Cosmology and

The KIAA-Cambridge Joint Workshop on Near-Field Cosmology and Galactic Archeology The LAMOST 1 d Spectroscopic Pipeline A-Li LUO LAMOST team, NAOC 2008/12/3

Lessons from SDSS • Three 1 d pipelines of SDSS (template based ) Princeton 1 d; Fermi 1 d; SEGUE: SSPP • Have been improving from DR 1 ->DR 7

Task of 1 D pipeline • Classification and Identification • Measurement (z of galaxies and QSOs, rv of stars) • Stellar parameter estimation • Special Candidate searching (Supernovae, Metal-poor stars, HII …) – according to requirements of astronomers

Software Structure Measurement Classification Preprocessing Image processing & Modular Spectra extraction ODBC/JDBC Interface QL DBMS File Management System Database Management Interface Storage &distribution CCD Raw Data

Production 1. Catalogs 2. Calibrated spectra with analysis results AGN Galaxies (z) Normal galaxies Stars QSOs Input Catalog Starburst Stars (rv) Emission Line stars O B Stars QSOs (z) A F G K Stars M or later Stars Unknown Results: Reflection Nebulae Basic Production Reference classification Supernovae Search H II Identification Stellar. Atmospheric. Parameters Multi. Wavelength Identification Candidate Catalogue

Comparison between object type and spectral class in SDSS DR 5 UNKNOWN STAR GALALY QSO HIZ_QSO STAR -LATE Object type Total number Correct (after spectral Identify) False (after spectral Identify) QSO 112147 58562 52. 219% 53585 47. 781% 3077 2. 744% 22975 20. 487% 16716 17. 905% 53855 48. 022% 4707 4. 197% 10817 9. 645% GALAXY 565267 548789 97. 085% 16478 2. 915% 3403 0. 602% 7333 1. 297% 548789 97. 085% 1179 0. 209% 3 0. 00053% 4560 0. 807% STAR 29595 28991 97. 959% 604 2. 04% 175 0. 591% 13426 45. 366% 52 0. 176% 217 0. 733% 160 0. 54% 15565 52. 593% -- object type -- spectral class

Classification algorithm • Automated Classification by objective methods (training by templates, predicting by distance or density ), collaborators: IA(CAS), BNU, SDU, etc. • Identified by line measurement

Identification automatically Extracted Spectra Absorption band detection Yes Late type stars (M type) with bands (Ti. O etc) No Emission line at 6563± 20 A, 4860± 20 A, 4340± 20 A ? Continuum fitting No No line spectra Lines detection Emission Line Spectra ? No No BL LAC or high Z galaxies Starburst AGN or QSO etc. Yes S/N low? No Continuum High or low ? Low Redshift measurement Yes High H II Region O_III 5007, H_alpha H_beta Absorption lines of Absorption lines at 6563± 20 A, 4860± 20 A, 4340± 20 A ? No NII 6583 measurement Na. I, Mgb and Ca. II etc He II lines No Normal galaxies Yes O or early B type star Yes A, F, G, early K star or Reflection Nebular Yes Late type emission line star + CSM No Early type emission line star + CSM Star forming galaxies Star burst galaxies Seyfert II QSO & Seyfert I LINER

STELLAR ANALYSIS PIPELINE A, F, G, K type stellar spectra Continuum Rectification Rough model spectra grid Teff~500 K, logg~1. 0 dex, [Fe/H]~1. 0 dex Line index definition H_delta, H_zeta, , Ca. II triplet, H&K, G band BAD Health Check? Sub-grid model spectra Teff~100 K, logg~0. 25 dex [Fe/H]~0. 25 dex Best fit rough spectra Cross-correlation Vrad geo Correction Cross-correlation Best fit spectra ± 10 -20 km/s Line Index Measure GOOD Line index & Color index calibration (ANN, Polynomial) Optimization of different methods Color index from Input Catalog [Fe/H] [C/Fe] Teff logg Absolute Magnitude trash bin High Resolution Spectra for example. HERES: 372 stars (VLT/UVES) R=20000 S/N=50 Visual Magnitude distance

Line Indices • To determine the local continuum level • Width selection

Some lines used in the pipeline • • • Ca. II K line (3933 A) Balmer lines Ca. II triplet Mg I b G band [C/Fe] Colors

Ca. II K ~ [Fe/H] Relationship between [Fe/H] and Ca. II K in 4500 K, 5000 K, 5500 K, 6000 K, 6500 K, 7000 K and 7500 K respectively (Marcs model synthetic spectra). Lines (left) and 2 order polynomial (right) are used to fit the relationships from low to high temperature. Relationships between [Fe/H] and the strength of Ca. II K in SDSS/SEGUE (Dr 6).

Balmer lines ~ Teff Three Balmer lines in Kurucz model spectra Hγ (434. 0 nm) [Fe/H] =-3. 0 [Fe/H] =-1. 0 [Fe/H] =-2. 0 [Fe/H] =0 Hδ (410. 2 nm) [Fe/H]=-3. 0 [Fe/H]=-1. 0 [Fe/H]=-2. 0 [Fe/H]=0 Hζ (388. 9 nm) [Fe/H]=-3. 0 [Fe/H]=-1. 0 [Fe/H]=-2. 0 [Fe/H]=0 • Hδ and Hζ in CFLIB spectra are obvious correlated with Teff. • Since the resolution of 1 Å FWHM of CFLIB and low S/N in the range around Hζ for half of the CFLIB dataset, Hζ line in 3889 Å is difficult to measure. • Fitting Teff ~ Hδ : Teff = 4572. 813 + 546. 716×Hδ − 53. 773×Hδ 2 error: 100 -200 K

Ca. II triplet Fitting of relationship between Ca. II triplet and Teff, [Fe/H], and logg respectively, CFLIB spectra were used as experimental dataset Relationship between Ca. II triplet and [Fe/H], EW of all Ca II triplet of SDSS/SEGUE spectra are plotted in left panel, and [Fe/H] varies with Ca. II triplet when T = 5000 K, logg = 2. 0 in right panel.

Mg. I b ~ gravity (left) SDSS data, (right) ELODIE data.

G band ~ [C/Fe] Relationship between G band [C/Fe] with HES follow up spectra

Color ~ Teff Temperature varies with B-V Color in CFLIB dataset For SDSS, in the range -0. 3 < g-r <1. 0, the following expression provides the effective temperature with an rms only 2% (100 -200 K) (Ivezić et al 2006)

Structure of the stellar analysis pipeline Independent compiled module +script Already completed module list: Kurucz model Continuum fitting ANN calculation (whole range) Module Spectra synthesize Continuum fitting Interpolation (local range) module Cross correlation Line index calculation Regression module EW calculation module

Kurucz model calculation • Atlas 9 Kurucz/Castelli • LTE • New. ODF • Intermod: an interpolation program to quickly generate intermediate models from an initial grid Spectra Synthesize • Synthe • Spectrum Gray

Test with Elodie library

Accuracy of the parameters • Checked with SEGUE dr 6 data

Accuracy of parameters with different SNR

Thanks !