紅移<1.5具明顯恆星誕生的低質量星系

全文

(1)國立臺灣師範大學理學院地球科學系研究所碩士論文 Department of Earth Science College of Science. National Taiwan Normal University Master Thesis. 紅移 ≤1.5 具明顯恆星誕生的低質量星系 Low-mass galaxies with extremely active star-formation at z ≤ 1.5. 林征發 Chen-Fatt, Lim. 指導教授：橋本康弘 & 傅谷石 Advisor: Yasuhiro Hashimoto & Sébastien Foucaud. 中華民國 104 年 8 月 August, 2015.

(2) 國立臺灣師範大學地球科學系研究所. 碩士論文. 紅移 ≤1.5 具明顯恆星誕生的低質量星系. 林征發. 104 8.

(3) 致謝由衷的感謝我的指導教授傅谷石老師（Seb）以及橋本康弘老師（Yas）。在老師們悉心的指導下，我才能完成這篇論文。從製定論文的大方向、技術問題以及最後的結果討論，各方面都離不開老師們耐心的幫助和教導。從老師們的身上也學到了做研究認真的態度與做學問精確嚴謹的方法。不僅如此，老師們還經常將他們的人生經歷以及待人處事的態度分享與我，讓我受益良多，也為我今後的工作樹立了優秀的榜樣。在此也特別感謝陳林文老師。雖然不是陳老師的指導學生，陳老師卻毫不吝嗇的解答我所遇到的問題和困難，以及給予我許多寶貴的意見。此外，也要感謝平下博之老師（Hiro）很細心的幫我改正論文，以及給了我許多珍貴的建議。此外，我還要感謝天文組的所有成員，所有幫助過我的學長姊、學弟妹以及同學們。特別感謝斌威、冠州、經閔、蕭赫、翔宇學長；明儀、安理、幼玲學姊，他們的樂於指導以及無私的幫助。感謝同研究室的陳寬、守倫、志鈞、嘉儒；同學姿穎、鴻選、宗賢，讓我在課業、研究以及生活上都得到許多的幫忙。當然還要感謝妳－惠頎，給了我很多的鼓勵與支持，在我心情煩悶時給我關懷與快樂。最後還要感謝我的家人，感謝你們一直以來對我的關心與照顧，有了你們無私的支持與關懷才讓我能夠專注於課業與研究。征發於 2015 年 6 月 25 日. i.

(4) 中文. 要. 我們發展出了一套基於測光系統的方法，用來分離出一種相對低質量但活躍於恆星誕生的星系在紅移 ≤1.5 的天區範圍內。我們將此方法運用在英國紅外望遠鏡 (United Kingdom InfraRed Telescope, UKIRT) 的紅外深度觀測 (UKIDSS-Ultra Deep Survey)，加法夏望遠鏡 (CanadaFrance-Hawaii Telescope, CFHT) 的勘測 (Legacy Survey) 以及宇宙學演化觀測 (COSMOlogical evolution Survey, COSMOS) 中的多波段探測資料 (Ultra Visible and Infrared Survey Telescope for Astronomy- UltraVISTA)。我們也結合了各個多波段探測的光譜觀測資料，得到了這類型星系的細微特徵。這類型的星系具有相對低的恆星質量 (stellar mass, M∗ ∼109 M⊙ )，低金屬豐度 (metallicity, 12+log(O/H)∼8.0)；但擁有高的絕對恆星誕生率 (specific star formation rate, SSFR∼10−9 M⊙ yr−1 /M⊙ )。這類型的星系也相似於靠近我們的藍色致密矮星系 (Blue Compact Dwarfs, BCDs)，在紅移 ∼0.8 的低元素豐度但是具有強恆星誕生的星系 (luminous metal-poor star forming galaxies) 以及為在高紅移的萊曼 α 發射體 (Lyα Emitters, LAEs)。關. : 星系: 矮星系 – 星系: 演化 – 星系: 恆星誕生 – 星系: 星爆 –. 星系: 高紅移. ii.

(5) Abstract We have recently developed a method based on broad-band photometry to isolate a sample of galaxies with relatively low stellar mass but extremely active star-formation at redshift up to z∼1.5. We extend our method to select those galaxies from several deep photometry archive datasets which are the UKIRT infrared deep sky survey - Ultra Deep Survey (UKIDSS-UDS), Canada-France-Hawaii Telescope Legacy Survey (CFHTLS) and COSMOSUltraVISTA survey. The spectroscopic data for each of the catalogs give us a good chance to look at the properties of our samples. These galaxies have low stellar mass (M∗ ∼109 M⊙ ) and low metallicity (12+log(O/H)∼8.0); however, they display high star formation rate per stellar mass (SSFR∼10−9 M⊙ yr−1 /M⊙ ). Interaction-induced infall of gas could be consistent with these galaxies that high specific star formation rates. These galxies may allow to study in great detail processes, such as starburst activity and chemical enrichment, which are common in the early universe. The properties of low metallicity for these galaxies are similar with Blue Compact Dwarfs (BCDs) in the local Universe, luminous metal-poor star forming galaxies at intermediate redshifts at z∼0.8 and Lyα Emitters (LAEs) at high redshifts. We will discuss more details about our photometic selection and the properties of these young galaxies. Key words: galaxies: dwarf – galaxies: evolution – galaxies: star formation – galaxies: starburst – galaxies: high-redshift. iii.

(6) Contents 致謝中文. i 要. ii. Abstract. iii. Contents. iv. List of Figures. vi. List of Tables. vii. 1 Introduction. 1. 2 Photometric Data. 3. 2.1. UKIDSS Ultra Deep Survey . . . . . . . . . . . . . . . . . . . . . . . .. 3. 2.2. CFHT Legacy Survey (deep field) + Palomar Near-IR catalog . . . . . .. 4. 2.3. COSMOS + UltraVISTA catalog . . . . . . . . . . . . . . . . . . . . . .. 4. 2.4. Conversion of Magnitude among Filters . . . . . . . . . . . . . . . . . .. 4. 3 Peas Selection Criteria. 8. 4 Spectroscopic Data. 11. 5 Properties of Emission Line Galaxies. 20. 5.1. “Baldwin, Phillips & Terlevich” (BPT) - diagram . . . . . . . . . . . . .. 21. 5.2. Stellar Mass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 22. iv.

(7) 5.3. Star Formation Rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 24. 5.4. Metallicity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 26. 6 Results and Discussions. 28. 6.1. Stellar Mass versus SFR . . . . . . . . . . . . . . . . . . . . . . . . . .. 28. 6.2. Stellar Mass versus SSFR . . . . . . . . . . . . . . . . . . . . . . . . . .. 28. 6.3. Mass-metallicity Relation (MZR) . . . . . . . . . . . . . . . . . . . . . .. 30. 6.4. Luminosity-metallicity Relation (LZR) . . . . . . . . . . . . . . . . . . .. 34. 6.5. Redshift versus SFR Indicators Luminosity . . . . . . . . . . . . . . . .. 36. 7 Summary. 39. A Conversions between Photometric Systems. 42. References. 44. v.

(8) List of Figures 1. Stars color-color diagram . . . . . . . . . . . . . . . . . . . . . . . . . .. 7. 2. SDSS- and NTT-peas spectra . . . . . . . . . . . . . . . . . . . . . . . .. 10. 3. UDS color-color diagram . . . . . . . . . . . . . . . . . . . . . . . . . .. 15. 4. CFHTLS color-color diagram . . . . . . . . . . . . . . . . . . . . . . . .. 17. 5. COSMOS color-color diagram . . . . . . . . . . . . . . . . . . . . . . .. 19. 6. EWs (Ours) versus EWs (Cardamone) . . . . . . . . . . . . . . . . . . .. 21. 7. BPT diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 23. 8. Ks-band Absolute Magnitude versus stellar mass . . . . . . . . . . . . .. 25. 9. Mstar versus SFR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 29. 10. Mstar versus SSFR . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 31. 11. SSFR histogram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 32. 12. SSFR CDF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 33. 13. Mstar versus Metallicity(MZR) . . . . . . . . . . . . . . . . . . . . . . .. 35. 14. g-band Absolute Magnitude versus Metalllicity(LZR) . . . . . . . . . . .. 36. 15. Redshift versus SFR indicators luminosity . . . . . . . . . . . . . . . . .. 38. vi.

(9) List of Tables 2.1. The list of the photometric data . . . . . . . . . . . . . . . . . . . . . . .. 6. 4.1. The summary of our photometric criteria . . . . . . . . . . . . . . . . . .. 13. vii.

(10) Chapter 1 Introduction The observed star-formation law indicates that stars form in cool gas reservoir such as molecular clouds (e.g., Kennicutt 1998; Bigiel et al. 2008; Daddi et al. 2010b). However the star formation law of young galaxies, especially at high-redshift is a matter of debate. We also have evidence that the star formation rate density (SFRD) declines steeply from z∼2 to the present and that the locus of star-formation migrates from massive and dense regions in the Universe to low-mass and isolated galaxies (Cowie et. al., 1996). Visible spectroscopic observations for low reshift star-forming galaxies are frequently observed and determined the detail about their chemical composition (Liang et al. 2006, Itozov et al. 2013). Cardamone et al. (2009) identified a class of rare and compact galaxies that display emission lines with extremely high equivalent widths at z∼0.25 based on their unusually vivid green appearance (r-i ≤ - 0.2 and g-r ≥ r-i + 0.5) in the SDSS g,r,i images (so called them “green pea” galaxies). The “green pea” galaxies have low-stellar mass(∼109.5 M⊙ ), low metallicities(log[O/H]+12∼8.7), low reddening(E[B-V] < 0.25) and reside in low density environments but have high star formation rates(∼10M⊙ yr−1 ). The high resolution HST images reveal that the “green pea” galaxies exhibit a disturbed optical morphologies. More details about the chemical abundances of “green pea” galaxies are examined by Amorín et al. (2010). The metallicity for all the “green pea” galaxies from Cardamone et al. (2009) are re-estimated by the reliable measurement of the [OIII]λ4363/ (λ4959+λ5007) emission lines (Te-method) and the strong-line calibrator [NII]λ6584/Hα 1.

(11) (PMC09-method, Pérez-Montero & Contini 2009). They find that the “green pea” galaxies have lower metallicities (log[O/H]+12∼8.0) compared with the estimation from Cardamone et al. (2009). The offset of mass-metallicity relation represent a ≥ 0.3 dex to lower metallicities. They argue that recent interaction-induced inflow of gas, possibly coupled with a selective metal-rich gas loss, driven by supernova winds, may explain the properties of “green pea” galaxies which is high specific star formation rates, extreme compactness and disturbed optical morphologies. Following the photometric selection criteria used by Cardamone et al. (2009) and combining the spectroscopic data from SDSS-DR7, Izotov et al. (2011a) studied a larger sample of 803 star forming Luminous Compact Galaxies (LCGs) which have properties similar to “green pea” galaxies. They argue that the “green pea” galaxies are not a specific class of objects but are just a subset of LCGs. The LCGs have wider redshift distribution (z∼0.02-0.63) than “green pea” galaxies, the median stellar mass is ∼109 M⊙ and star formation rate varies in a large range of 0.7-60M⊙ yr−1 with a median value of ∼4M⊙ yr−1 . Ultilizing the reliable measurement of [OIII]λ4363/(λ4959+λ5007) emission lines (Temethod), they find that LCGs have a lower median value of metallicity (log[O/H]+12∼ 8.11) compare with the result of Cardamone et al. (2009), confirming these objects are low in metallicity. In this thesis, we investigate the properties, such as stellar mass, star-formation rate and chemical abundances, about these relatively isolate but active in star-formation galaxies from different fields. We develope a method based on the broad-band photometry to select similar objects to “green pea” galaxies and LCGs in some various redshift ranges up to z∼1.5 from different fields. Our photometric data is described in Chapter 2. The selection criteria of our galaxies are explained in Chapter 3. The sepctroscopic data we used is discussed in Chapter 4. In Chapter 5, we describe the properties about our galaxies samples. In Chapter 6, we discuss the physical properties for our sample galaxies. We summarize our results in Chapter 7. Through the paper, we assume a ΛCDM cosmology with Ωm = 0.3, ΩΛ = 0.7, H0 = 70kms−1 Mpc−1 .. 2.

(12) Chapter 2 Photometric Data Photometry is a technique of astronomy concerned with measuring the flux which inside a particular region of the sky. When photometry is done with a set of broad wavelength bands of instrument, not only the amount of radiation but also the spectrum distribution is measured. The photometric data released from different surveys can provide us a plenty of valuable information about the stellar (or galactic) spectral distribution. The photometric data in this thesis are shown as below.. 2.1 UKIDSS Ultra Deep Survey United Kingdom InfraRed Telescope (UKIRT) is a 3.8 metre infrared reflecting telescope which located on Hawaii. A near-IR Wide Field CAMera (WFCAM) has been designed specifically to carry out large scale survey observations for UKIRT. Ultilizing the WFCAM, a near-IR sky survey UKIRT Infrared Deep Sky Survey (UKIDSS) was begun in May 2005 and will survey 7500 square degrees of the Northern sky. The Ultra Deep Survey (UDS-DR10) is the deepest component of the UKIDSS. The UDS covers an area of 0.8 square degrees, which centred on the Subaru-XMM Deep field (J0218-05), with a median depth of UKIRT K=25.0 (5σ, AB). The UDS catalog provides a wealth of multiwavelength data in this field which includes the XMM-Newton, CFHT Megacam, Subaru Suprime-cam, UKIRT WFCAM, VISTA, SPITZER IRAC/MIPS, Herschel SPIRE, SCUBA/AzTEC, VLA and HST WFC3. In this catalog, we focus on Subaru 3.

(13) Suprime-cam(B, V, Rc, i’, z’ bands) and UKIRT WFCAM(J, H, K bands).. 2.2 CFHT Legacy Survey (deep field) + Palomar Near-IR catalog The Canada-France-Hawaii Telescope Legacy Survey (CFHTLS) is based on a dataset collected with MegaCam on the CFHT. The CFHTLS-Deep field has four independent 1 square degrees MegaCam ultra deep pointings, providing a high quality datset in five optical bands (u*, g, r, i, z bands). To supplement near-IR data, we adopted AEGIS: Palomar Near-IR photometric catalogs (J-, Ks-bands) data and concentrated on the CFHTLS-D3 dataset which shares the same field as the AEGIS.. 2.3 COSMOS + UltraVISTA catalog The UltraVISTA survey (McCracken et al., 2012) has imaging data in four near-IR broad bands (Y, J, H, Ks bands) and covers the same field as the COSMOS survey. The extended COSMOS/UltraVISTA is a Ks-selected catalog of the COSMOS field. Catalog sources are selected from the DR1 UltraVISTA Ks- band imaging that reaches a depth of 23.4 AB (90% completeness). The COSMOS+UltraVISTA catalog covers a total area of 1.62 square degrees and 30 photometric bands with PSF-matched photometric data including the GALEX, Subaru, CFHT, UltraVISTA and Spitzer imaging are available. In this catalog, we focus on the Subaru Suprime-cam (B, V, r’, i’, z’ bands) and the UltraVISTA VIRCAM (J, H, Ks bands) which provide us a plenty of good quality photometric data from opitcal to near-IR wavelengths.. 2.4 Conversion of Magnitude among Filters For convenience, all the magnitudes in optical and NIR filter sets in each catalog are converted to those of the SDSS(u, g, r, i, z) and UltraVISTA(J, H, Ks) filters. The formulae. 4.

(14) for the conversions are given in Appendix. To test our calibration, we extract the information of main sequence stars from UDS (star flag = 1), CFHTLS (class star > 0.95 and g < 23) and COSMOS (type = 1.0) then adopt their photometric observations into several color-color diagrams. The position of star sequence in the gri- and gzKs- diagram (Fig. 1) nicely overlap after applying the magnitude conversions. We also apply the SDSS spectroscopically confirmed stars into the gri- diagram (red points in figure 1). Due to the lack of K-band information, SDSS stars are not included in the gzKs- diagram. Although some small offsets (∼0.1 mag) still exist on the color-color diagrams between the catalogs, those offsets do not affect our results significantly.. 5.

(15) Table 2.1: The list of the photometric data Photometric Catalog. Instrument. Subaru Suprime-Cam UKIDSS UDS. UKIRT WFCAM. CFHTLS Mega-Cam CFHTLS + Palomar Palomar Near-IR. Subaru Suprime-Cam COSMOS+UltraVISTA. UltraVISTA VIRCAM. 6. Filter B V Rc i’ z’ J H K u* g r i z J Ks B V r’ i’ z’ J H Ks. λeﬀ (nm) 447.8 549.3 655.0 770.9 905.4 1250.0 1630.0 2200.0 375.0 485.0 625.0 770.0 885.0 1250.0 2140.0 447.8 549.3 631.5 770.9 905.4 1250.0 1650.0 2150.0.

(16) Figure 1: (a) g-r versus r-i color-color diagram of stars in all the datasets adopted. Light blue, green and black color points represent the UDS, CFHTLS and UltraVISTA samples respectively. Because the final results of our visible filter systems are in the SDSS base, we add the SDSS stars observation in this diagram (red points). (b) g-z versus z-Ks diagram of stars from all of our datasets. Due to the lack of K-band information, SDSS stars are not included in this diagram. The stars occupy similar positions in the color-color diagrams, which indicates that our filter transformations are reliable. 7.

(17) Chapter 3 Peas Selection Criteria To define our selection criteria, we use 80 low-redshift SDSS peas spectra (z∼0.3, Cardamone et al., 2009) and 8 higher-redshift ESO-NTT pea spectra (z∼0.6, Bamford et al., 2014) as spectrum templates. The spectra of the 80 SDSS peas were taken from SDSS-III Science Archive Server which contains a well co-added calibrated flux measured through a fiber with 3” circular aperture. The spectra of 8 higher-redshift NTT-peas are taken from the NTT EFORSC2 intrument by a grism with 1.2” slit width. Convolving these spectrum templates with broad-band photometric system enables us to define the peas selection criteria in color-color diagrams. The coverages of spectroscopic data for SDSS and NTT are not enough to convolve with the broad-band systems; therefore, we use their individual photometric information taken from the SDSS and UDS photometric catalog (3” aperture for SDSS and 2” aperture for NTT) to extend our spectral templates. To make a fair comparison between photometric and spectroscopic data, we apply the aperture correction into 2” for NTT spectra flux by multiplying the spectral flux by 1+(1.2”/2”)2 . This factor is coming from the ratio of aperture areas that between the photometric and spectroscopic data. In the case of slits (or fibres), small slit width (or aperture size) will physically lose flux compare with the bigger one due to they have different size of physical aperture. Therefore, in our case, we need to apply the correction into the relative smaller slit width of NTT spectroscopic data and make them consistent with their photometric data. Figure 2 represents the example of spectroscopic data for SDSS and NTT pea. The photometric flux estimated from the 8.

(18) above procedure are also shown as red-crossed points in Figure 2. In order to test our convolution, we employ those extended SDSS/NTT peas spectra to convolve them with the response of SDSS/UDS filters in original redshift. Then, we compare the results with their own photometric observation. Our results show a good consistency within photometric and spectroscopic data (△magnitude ∼ 0.05 mag). These 88 spectral templates are shifted to our specific redshift ranges of interest (z=0.150.40, z=0.40-0.65, z=0.65-1.20 and z=1.35-1.65) with a redshift interval of 0.01 and we convolved the spectra with the response of SDSS (optical) and UltraVISTA (NIR) filters to predict the magnitude of our candidates in each broad-band filter. They cover a similar region (green contours in Fig. 3, 4, 5) in several color-color diagrams such as g-r versus r-i, r-i versus i-z, i-z versus z-J and z-J versus J-Ks. The similar region in color-color diagrams will enable us to identify the objects that unusually luminous in r-, i-, z- and J- bands. The redshift interval of 0.01 will allow us to define the finest distribution for “green pea” galaxies in those color-color diagrams. Thanks to the deep photometric datasets available (UDS-DR10, CFHTLS+Palomar and COSMOS+UltraVISTA), we can extend our broad-band method to select the r-, i-, z-, J- luminous objects which could be the condidates of “pea galaxies” at z∼0.25, z∼0.55, z∼0.85 and z∼1.5. Note here that almost all of our color-color diagrams are shown as the subtraction of nearest photometric systems except our J-selection. We take the Ks-band instead of Hband in z-J versus J-Ks color-color diagram due to the lack of H-band data in the CFHTLS +Palomar catalog. Beside that, we also show the location of SDSS-peas in g-r versus r-i color-color diagram ( green star symbols in Figs. 3a, 4a, 5a).. 9.

(19) Figure 2: (a) One sample from our SDSS spectrum template. Red cross points represent the SDSS u, g, r, i, z phomotometric information which correspond to their effictive wavelength respectively. (b) Same as in (a) but sample from our NTT spectra template. The red cross points represent the UDS U, B, V, Rc, i, z photometric information.. 10.

(20) Chapter 4 Spectroscopic Data To further investigate the properties of our samples, we need spectroscopic data for each of the broad-band surveys. For the UDS dataset, an UDSz ESO Large Programme (180.A-0776, PI: Almaini) which provides targeted spectral observations for ∼3000 Kband selected galaxies, was completed using VIMOS and FORS2. Galaxies were selected over 0.6 square degrees of the UDS field to a limit of K=23(AB). The spectroscopy of the CFHTLS+Palomar dataset were carried out utilizing the DEEP2 DEIMOS (DEep Imaging Multi-Object Spectrograph) on the Keck II 10m telescope (Newman et al. 2013). DEEP2 Galaxy Redshift survey targeted almost 50,000 galaxies at z < 1.4. Finally, the zCOSMOS redshift survey is performed by VIMOS, providing approximately 20,000 i-band selected galaxies at redshifts z < 1.2 in 1.7 square degrees of the COSMOS field (Lilly et al. 2007). The spectroscopic data in each of color-color diagrams can be ultilized to comfirm the color selection for our data. Therefore, we add all the spectroscopic data into the color-color diagrams (big points in Figs. 3, 4 and 5). For convenience, we also separate the spectroscopic data into emission line galaxies (flag 2, number of emission line ≥ 2), potential emission line galaxies (flag 1, number of emission line=1) and not emission line galaxies (flag 0). The emission lines flag information associate with redshift distribution can made us to select the emission line galaxies and potential emission line galaxies with expected redshift from the entire samples (big blue and light blue points in Figs. 3, 4 and 5). The distribution of emission line galaxies and potential emission line galaxies with different redshift also represent in our color-color diagrams (big yellow and purple points 11.

(21) in Figs. 3, 4 and 5). Beside that, we also separate our spectroscopic data by three different photometric criteria which are loose-, middle- and tight-criterion (small red, orange and green points in Figs. 3, 4 and 5). Loose-criterion almost can include all the emission line galaxies with our expected redshift in color-color diagrams. In order to extract the candidates of “green pea” galaxies from our catalogs, we divide our spectroscopic data into two more categories with middle- and tight-criterion, which are more consistent with the distribution of “green pea” galaxies in color-color diagrams (green contours in Figs. 3, 4 and 5). More details about our photometric criteria are described in Table 4.1.. 12.

(22) Table 4.1: The summary of our photometric criteria loose-criterion(a). middle-criterion(b). r-selection. (r - i) < 0.5 × (g - r) + 0.05 and (g - r) > 0.1. (r - i) < 0.15 and (g - r) > 0.2. tight-criterion(c) (r - i) < 0.42×(g - r) - 0.18 and (r - i) < 0.15 and (g - r) > 0.2. i-selection. (i - z) < 0.75×(r - i) - 0.025 and (r - i) > 0.1. (i - z) < 0.75×(r - i) - 0.025 and (r - i) > 0.2 and (i - z) < 0.2. (i - z) < 0.75×(r - i) - 0.25 and (r - i) > 0.2 and (i - z) < 0.2. z-selection. (z - J) < 0.65×(i - z) + 0.33 and (i - z) > 0. (z - J) < 0.2 and (i - z) > 0.1. (z - J) < 0.5 × (i - z) - 0.15 and (z - J) < 0.2 and (i - z) > 0.1. J-selection. (J - Ks) < 0.81×(z - J) + 0.11 and (z - J) > 0.2. (J - Ks) < 0.81×(z - J) + 0.11 and (z - J) > 0.3 and (J - Ks) < 0.7. (J - Ks) < 0.66 ×(z - J) - 0.33 and (z - J) > 0.3. (a) Loose-criterion almost can include all the emission line galaxies with our expected redshift in color-color diagrams. In order to extract the candidates of “green pea” galaxies from our catalogs, we divide our spectroscopic data into two more categories with (b) middle- and (c) tight-criterion, which are more consistent with the distribution of “green pea” galaxies in color-color diagrams (green contours in Figs. 3, 4 and 5).. 13.

(23) 14.

(24) Figure 3: Color-color diagram for the UDS catalog. Small and big points correspond to the UDS photometric and spectroscopic data respectively. Blue and light blue big points represent the emission line galaxies (ELGs) and potential emission line galaxies at our expected redshift. Yellow and purple big points represent the emission line galaxies and potential emission line galaxies but out the redshift range. Red, orange and green small points indicate the loose-, middle- and tight- photometric criteria which are discribed in Section 4. Green contours correspond to the distribution of our 88 pea spectral templates. (a) r-i versus g-r diagram, aiming at selecting the 0.15<z<0.35 pea galaxies. Green star symbols represent the location of SDSS-peas galaxies in g-r versus r-i color-color diagram. (b) i-z versus r-i diagram, aim to select the 0.4<z<0.7 pea galaxies. (c) z-J versus i-z 15 diagram, aim to select the 0.7<z<1.2 pea galaxies. (d) J-Ks versus z-J diagram, aim to select the 1.3<z<1.7 pea galaxies..

(25) 16.

(26) Figure 4: Same as Fig 3. but for the CFHT-LS sample. Small and big points correspond to the CFHT-LS photometric and spectroscopic data respectively. (a) r-i versus g-r diagram, aim to select the 0.15<z<0.35 pea galaxies. Green star symbols represent the location of SDSS-peas galaxies in g-r versus r-i color-color diagram. (b) i-z versus r-i diagram, aim to select the 0.4<z<0.7 pea galaxies. (c) z-J versus i-z diagram, aim to select the 0.7<z<1.2 pea galaxies. (d) J-Ks versus z-J diagram, aim to select the 1.3<z<1.7 pea galaxies.. 17.

(27) 18.

(28) Figure 5: Same as Fig 3. but for the COSMOS sample. Small and big points correspond to the COSMOS photometric and spectroscopic data respectively. (a) r-i versus g-r diagram, aim to select the 0.15<z<0.35 pea galaxies. Green star symbols represent the location of SDSS-peas galaxies in g-r versus r-i color-color diagram. (b) i-z versus r-i diagram, aim to select the 0.4<z<0.7 pea galaxies. (c) z-J versus i-z diagram, aim to select the 0.7<z<1.2 pea galaxies. (d) J-Ks versus z-J diagram, aim to select the 1.3<z<1.7 pea galaxies.. 19.

(29) Chapter 5 Properties of Emission Line Galaxies The spectra of emission line galaxies at different redshifts in those different fields could tell us many details about the properties of these galaxies. Here, we only focus on the true emission line galaxies sample (flag 2) for the r- and i- selection. However, both true and potential emission line galaxies (flag2 and flag 1) are considered for the z- and J-selection because the narrow wavelength coverage of spectra often allows us to identify one emission line at most. In order to estimate the luminosity and equivalent width of the emission lines, we adopted the IDL routine - gaussfits. First we exclude all the emission lines from the spectra respectively to estimate their continuum flux level. We blue shift the spectra individually then exclude the rest-frame emission lines by the ±15 Å. After that, we deduct the continuum flux from the original spectra and use the IDL routine - gaussfits to fit the emission line in gaussian at a certain range of spectra. We can derive the luminosity of emission line by integrating the gaussian fitted to the emission line profile. The equivalent width of emission line is also derived by their luminosity and continnum flux level. The spectral resolution may play an important role in our flux or equivalent width estimation. The spectral resolution of our spectroscopic data are represented as UDSz ∼ 6Å (VIMOS ∼ 5.5Å and FORS2 ∼ 6.4Å), DEEP2 DEIMOS ∼ 1.6Å and zCOSMOS ∼ 5.5Å. The emission lines width of green peas galaxies are ∼ 25Å. Even though the worst case of spectral resolution (∼ 6.4Å) for UDSz data, the emission lines can be resolved and fitted by Gaussian. We claim that the spectral resolution for each of the spectroscopic data 20.

(30) Figure 6: The comparison of our [OIII]λ5007Å equivalent width estimation and [OIII]λ5007Å equivalent width estimated by GANDALF code for the SDSS-peas (table 4. of Cardamone et al. (2009) ). The solid line represents the equality of the two estimations. is enough for us to estimate their flux or equivalent width. Utilizing the equivalent width information of SDSS Green Peas galaxies (Cardamone et al. 2009), we can examine our equivalent width estimation. The [OIII]λ5007Å equivalent widths taken from table 4 of Cardamone et al. (2009) are derived from the SDSS database using the GANDALF(Gas AND Absorption Line Fitting) code. As shown in Fig. 6, our results are consistent with the GANDALF estimation, which indicates that our estimation is reliable.. 5.1 “Baldwin, Phillips & Terlevich” (BPT) - diagram The “Baldwin, Phillips & Terlevich” (BPT) - diagram is consisted of several sets of nebular emission line flux ratio which can distinguish the ionization mechanism of nebular gas come from star forming regions or AGNs. Although AGNs are already flagged in each of the catalogs, we still need to comfirm that there are free of AGNs feature in our emis21.

(31) sion line galaxies samples. Since the emission lines covered by spectroscopic observations could be different at different redshifts, only small fraction of our samples have both detected [OIII]λ5007Å/Hβ and [NII]λ6583Å/Hα in BPT-diagram (Fig. 7). However, most of the low redshift sample only have [OIII]λ5007Å/Hβ (or [NII]λ6583Å/Hα) detection which represents the horizontal (or vertical) histogram in Fig. 7. This empirical method succeeds in making the separation of AGNs and star-forming galaxies (Kewley et al. 2001, red curve in Fig. 7). The AGNs will exceed the red curve which will locate at upper, upper right and right in BPT-diagram. According to the value from Kewley et al. 2001, we filtered out 7 AGNs that exceed the red curve, 11 potential AGNs which represent the log([OIII]λ5007Å/Hβ) > 0.9 but found no potential AGNs for log([NII]λ6583Å/Hα) > 0.2. Unfortunately, our z-selection and all J- selection samples barely have [OIII]λ5007Å/ Hβ or [NII]λ6583Å/Hα detection; thus, we do not apply the BPT selection and just follow their AGNs flag information given by each of the catalogs.. 5.2 Stellar Mass Each of the catalogs has their own published stellar mass information. Photometric SED fitting provides stellar mass estimation for the UDS catalog (Coupon et al. 2013). Ultilizing optical color K-band and well-known spectroscopic redshift, the stellar mass information is published for the CFHTLS catalog which only contains galaxies with M∗ >1010 M⊙ (Bundy et al. 2006). Stellar masses for COSMOS catalog are derived by the best fit templates of the BC03 stellar population synthesis model (Ilbert et al. 2013). Theie stellar mass estimation may be overestimated by factor of 2 on average due to the contribution from emission lines will increase their photometric magnitude. From the relation between absolute magnitude and stellar mass shown in Fig. 8, we can interpole the information of stellar mass that M∗ <1010 M⊙ for CFHTLS catalog. We adopt the absolute magnitude of observed frame Ks-band in Fig. 8 which can avoid the effect from emission lines. Note here that we only apply the distance modulus for our Ksabsolute magnitude estimation without k-correction. The main reason is that we do not have more information about the redder part (>10000Å) in spectroscopic observation so 22.

(32) Figure 7: The BPT diagram of all emission line galaxies in the sample (flag 2 for r- and i-selection; flag 2 and flag 1 for z- and J-selection). Only small fraction of our nearby sample have both detected [OIII]λ5007Å/Hβ and [NII]λ6583Å/Hα. However, most of our samples still have one of the [OIII]λ5007Å/Hβ (or [NII]λ6583Å/Hα) detections, which are shown in the horizontal (or vertical) histogram. Circle, star and trianlge symbols represent the UDS, CFHTLS and COSMOS samples. Black, blue and green color represent the emission line galaxies selected from r-, i-, z-selection corresponding to z∼0.25, z∼0.55 and z∼0.85 respectively. The empirical line that separates AGNs and star-forming galaxies is shown by the red curve (Kewley et al. 2001). According to the value from Kewley et al. 2001, we filtered out 11 potential AGNs which represent the log([OIII]λ5007Å/Hβ) > 0.9 but found no potential AGNs for log([NII]λ6583Å/Hα) > 0.2. Unfortunately, our z-selection and all J- selection samples barely have [OIII]λ5007Å/Hβ or [NII]λ6583Å/ Hα detection; thus, we do not apply the BPT selection and just follow their AGNs flag information given by each of the catalogs.. 23.

(33) we barely have a well estimation of k-correction for our samples. The Ks- absolute magnitude of the UDS and COSMOS emission line galaxies correlate well with their stellar mass (circle and triangle symbols in Fig. 8). Variance slope of relation are mainly caused by the lack of k-correction in our samples, i.e. the observed frame of Ks-band represent the bluer part for the high redshift samples. The slope of Ks-stellar mass relation will become flat and no correlate if we adopt the IDL routine kcorrect (Blanton et al. 2007) into the Ks magnitude. We also exihibit this relation for the whole CFHTLS stellar mass catalog (Bundy et al. 2006. - red star symbols in Fig. 8). The relation from the CFHTLS catolog (red line in Fig. 8) is similar to that of the UDS and COSMOS samples at z∼0.55 and 0.95 (blue and green lines in Fig. 8). This is because the CFHTLS, which only includes the M∗ >1010 M⊙ samples, have a large fraction of high redshift samples. The median redshift of CFHTLS well-known stellar mass samples is z∼0.85. Ultilizing the individual relationship in their given redshift (black, blue and green color points in Fig. 8 correspond to z∼0.25, z∼0.55 and z∼0.95), we can estimate the robust information about the stellar mass for our CFHTLS catalog. In order to make a fair estimation, we only adopt the relation from CFHTLS (red line in Fig. 8) for our CFHTLS iand z-selection samples and use the relation from low-redshift UDS+COSMOS samples (black line in Fig. 8) to interpole the stellar mass information of our CFHTLS r-selection sample.. 5.3 Star Formation Rate The [OII]λ3727Åand Hα emission lines are known to be the good indicators of star formation rate (Kennicutt et al. 1998):. SF ROII [M⊙ yr−1 ] = 1.4 × 10−41 L[OII] [ergs−1 ]. (5.1). SF RHα [M⊙ yr−1 ] = 7.9 × 10−42 LHα [ergs−1 ]. (5.2). 24.

(34) Figure 8: Ks-band Absolute Magnitude versus Stellar Mass diagram. Circles, stars and trianlges represent the UDS, CFHTLS and COSMOS sample, respectively. Black, blue and green colors represent the emission line galaxies selected from the r-, i- and z-criteria corresponding to z∼0.25, z∼0.55 and z∼0.85, respectively. The red points symbolize the CFHTLS catalog (Bundy et al. 2006), which only have the information for samples that log(M∗ ) > 10. Variance slope of relation are mainly caused by the lack of k-correction in our samples, i.e. the observed frame of Ks-band represent the bluer part for the high redshift samples. In order to make a fair estimation, we only adopt the relation from CFHTLS (red line in Fig. 8) for our CFHTLS i- and z-selection samples and use the relation from low-redshift UDS+COSMOS samples (black line in Fig. 8) to interpole the stellar mass information of our CFHTLS r-selection sample.. 25.

(35) As explained above, the emission lines covered by spectroscopic observations could be different at different redshifts. Thus, we applied both SFR estimates for all of the emission line galaxies if there have both [OII]λ3727Å and Hα detections. In the case that we only have [OII]λ3727Å (or Hα) detection, we use equation (5.1) (or equation (5.2)) to derive their SFR. Filled and open symbols in Fig. 9 represent the SFRs derived from [OII]λ3727Åand Hα luminosity, respectively. A large fraction of our sample have the OII detection. In order to make a fair comparison, we only adopt the SFROII if our samples have both [OII]λ3727Å and Hα detections. The samples at z∼0.4-0.7 for the CFHTLS catalog do not have any [OII]λ3727Å or Hα detection, except that their spectroscopic observation cover the Hβ emission line. For these objects, we use the Hβ luminosity by assuming the theoretical ratio for Case B, L(Hβ)=L(Hα)/2.86, and adopt equation (5.2) to derived their SFR. Note here that we do not apply any extinction correction for the SFR estimates. Thus, we may be underestimated the SFR for our objects.. 5.4 Metallicity Metallicity among our samples are playing a key role for investigate their physical conditions and element abundances. The most direct method for measuring the metallicity is to determine the oxygen abandunce line ratio, such as I[OIII]λ4595,λ5007 /I[OIII]λ4363 (Temehod). Assuming a classical HII-region model, this line ratio provide an estimation of the electron temperature of the gas. Unfortunately, the temperature sensitive line [OIII]λ4363Å is always too weak to be observed and is not covered by the optical spectroscopic intrusments for high redshift samples (z > 1.0). Based on nearby galaxies, an emprical method (N2-method) were developed to determine the metallicity of high redshift samples (Pettini & Pagel et al. 2004):. 12 + log(O/H)N II = 8.9 + 0.59log(I[N II] /IHα ). (5.3). The (O/H) is the number ratio of oxygen to hydrogen atoms and I[N II] /IHα represents the intensity ratio of [NII]λ6583Åand Hα. However, this estimation should not be ap-. 26.

(36) plied at log(IN II /IH α ) > 0 for star-forming galaxies. Star-forming galaxies show a nearly constant value at log(IN II /IH α ) > 0 in BPT diagram (red curve in Fig. 7) which will lead to a constant value at 12+log(O/H) ∼ 8.9. Another empirical oxygen abundance indicator R23 = (I[OII]λ3727,λ3729 +I[OIII]λ4959,λ5007 )/ IH β (R23-method) has widely been used because the temperature sensitive line [OIII]λ4363Å is too weak to be detected (Pagel et al. 1979). A calibrated relation (Pilyugin et al. 2000) between (O/H) and R23 can be used to determine the metallicity of our samples; however, the results are double-valued:. 12 + log(O/H)R23 = 6.53 − 1.40log(R23 ). (5.4). (while 12+log(O/H)T e < 7.95). 12 + log(O/H)R23 = 9.50 − 1.40log(R23 ). (5.5). (while 12+log(O/H)T e > 8.15) Ultilizing the above equations, we can derive the metallicity for our emission line galaxies. Most of our samples have [OII], [OIII] and Hβ detection. To make a fair comparison, we first adopt the R23-method to derive the metallicity for our sample despite that it will lead to a double metallicity values (filled symbols in Figs. 13 and 14). We apply equation (5.3) to most of our low redshift samples which always lack [OII] data (open symbols in Figs. 13 and 14). Note here that we do not apply the N2-method to our UDS samples because the spectroscopic resolution of their intrusments - VIMOS and FORS2 are too low to seperate the [NII] and Hα emission lines. The results that take the assumption of 12+log(O/H)T e > 8.15 (equation 5.5) can be considered as upper limits for metallicity.. 27.

(37) Chapter 6 Results and Discussions 6.1 Stellar Mass versus SFR We represent the relation between stellar mass and SFR for the emission line galaxies which are selected from loose-, middle- and tight-criterion in color-color diagrams (Fig. 9). Black and green solid lines represent the relation SDSS sample at z∼ 0 and GOODS sample at z∼ 1 (Elbaz et al. 2007), respectively. Most of our objects are consistent with the stellar mass versus SFR relation from Elbaz et al. 2007. As discussed above in Section 5.2 and 5.3, the stellar mass and SFR may be overestimated and underestimated, respectively. Thus, the stellar-mass relation might be raised to upper left in the diagram. Emission line galaxies selected from our tight-criterion (Fig. 9c) could be the “green pea” galaxies and they show a big scatter in the SFR. However, comparing with the individual stellar mass information, the tight-criterion sample has higher SFRs at a given stellar mass, i.e. we efficiently select the upper left objects in stellar mass versus SFR diagram.. 6.2 Stellar Mass versus SSFR The specific star formation rate (SSFR), star formation rate normalized to the stellar mass, might be a good indicator for the efficiency of star formation in our samples. We present the relation between stellar mass and SSFR for our sample selected from loose-, 28.

(38) Figure 9: Relation between stellar mass and SFR for the emission line galaxies selected from the (a) loose-, (b) middle- and (c) tight- photometric criteria. Circles, stars and trianlges represent the UDS, CFHTLS and COSMOS samples. Black, blue, green and red indicate the emission line galaxies selected from r-, i-, z- and j-criteria which correspond to z∼0.25, z∼0.55, z∼0.85 and z∼1.5, respectively. We also include the NTT-peas which are shown in light blue. The filled symbols correspond to the SFR estimated by OII emission line and open symbols represent the SFR derived by Hα luminosity. Black and green lines represent the SFR-Mstar relation by SDSS 29 samples at z∼ 0 and GOODS samples at z∼ 1 (Elbaz et al. 2007)..

(39) middle- and tight-criterion (Figs. 10a, 10b and 10c). Dotted, dashed and dot-dashed lines in the figures correspond to the SFR = 1, 10, 100M⊙ yr−1 , respectively. We also show the histogram and cumulative distribution function (CDF) of SSFR for the samples selected by our loose-, middle- and tight-criterion (dotted, dashed and solid lines in Fig. 11 and Fig. 12). To make a fair comparison, we separate our emission line galaxies by their redshift to r-selection (z∼0.25), i-selection (z∼0.55) and z-selection (z∼0.95). The J-selected samples is not shown because of its small size. We also represent the CDF diagram for our entire samples which include our J-selected samples (Fig. 12d). The result of Kolmogorov–Smirnov test (K–S test) is also presented in our CDF diagrams (Fig. 12). P(LM), P(LT) and P(MT) represent the P value between loose-middle, loosetight and middle-tight samples respectively. The P value is computed from the maximum distance between the CDF. If the P value is smaller than 0.01, conclude that the two groups were sampled from populations with significantly different distributions. Regardless of the redshift range, the P(LT) shows that our tight-criterion samples are significantly different from the loose-criterion samples. This also indicates that our tight-criterion is very effective in selecting the high SSFR objects, .i.e. they have higher SFR at a given stellar mass compared with the emission line galaxies at similar redshift. Our tight-criterion sample (Fig. 10c) shares a similar range in stellar mass at different redshift; however, the SSFR is higher for higher-redshift objects. It is dangerous to claim any evolution among our samples because we cannot confirm that we are sampling the same population of galaxies at those different redshifts.. 6.3 Mass-metallicity Relation (MZR) Mass-metallicity relation (MZR) might tell us more detail about the star formation history and the evolution of metallicity among our sample galaxies. The solid lines in black, green, orange and red (Fig. 13) represent the mass-metallicity relation of z ∼ 0.07, 0.7, 2.2 and 3.5 from SDSS-DR2, UV selected LBGs from Keck I, Gemini Deep Deep Survey (GDDS) + Canada-France Redshift Survey (CFRS) and AMAZE sources respectively 30.

(40) Figure 10: Stellar Mass versus SSFR diagram. The symbols and colors are same as Fig. 9 panels (a), (b) and (c) show the sample selected by loose-, middle- and tight- photometric criteria respectively. Dotted, dashed and dot-dashed lines indicate the SFR = 1, 10, 100M⊙ yr−1 respectively.. 31.

(41) Figure 11: Histogram of SSFR. Dotted, dashed and solid lines correspond to the sample selected by our loose-, middle- and tight- photometric criteria. To make a fair comparison, we separate our emission line galaxies by their redshift that represent as (a) r-selection (z∼0.25) (b) i-selection (z∼0.55) and (c) z-selection (z∼0.85).. 32.

(42) Figure 12: Cumulative distribution function(CDF) of SSFR. The description of lines is same as Fig. 11. (a) r-selection (z∼0.25) (b) i-selection (z∼0.55), (c) z-selection (z∼0.85) and (d) the entire sample including the J-selection sample. The P(LM), P(LT) and P(MT) represent the P value of the Kolmogorov–Smirnov test (K–S test) for loose-middle, loosetight and middle-tight samples respectively.. 33.

(43) (Maiolino et al. 2008). As discussed above in Section 5.4, there are two possible values of metallicity if the R23-method is applied. We represent both metallicity results that are assumed 12+log(O/ H) T e < 7.95 (left panel) and 12+log(O/H) T e > 8.15 (right panel) in Fig. 13. The metallicity for most of our low redshift objects are derived by N2-method which lose the sensitivity to estimate the objects at 12+log(O/H) > 8.9 (open symbols in Fig. 13). Thus, the MZR for most of our low redshift objects has a “ceiling effect” at 12+log(O/H) T e ∼ 8.9. As we did in the previous subsection, we separate our samples with the loose-, middleand tight-criterion in Fig. 13 (top, middle and bottom panels). If we only look at the metallicity derived by R23-method, our tight-criterion objects are almost at the region around 12+log(O/H)T e ∼ 8.0 (filled symbols in two lowest panels of Fig. 13). This implies that regardless of taking the assumption of 12+log(O/H) T e < 7.95 or 12+log(O/H) T e > 8.15, the metallicity of the tight-criterion objects are at 12+log(O/H)T e ∼ 8.0. The MZR for the tight-criterion sample might be similar to the high redshift objects (z∼2.2, yellow trend in Fig. 13). Because of the double possible metallicity values, we cannot judge if there is a significant difference in metallicity among the loose-, middleand tight-criterion samples.. 6.4 Luminosity-metallicity Relation (LZR) We also show the relation between g-band absolute magnitude and metallicity for our samples (Fig. 14). For references, the relation from nearby emission line galaxies (Guseva et al. 2009) and luminous compact galaxies (LCGs) (Izotov et al. 2011a) are also shown in Fig. 14 which metallicity are derived by the direct method (Te-method). We adopt the IDL routine - kcorrect (Blanton et al. 2007) to derive the k-correction value and apply them into our g-band absolute magnitude. However, we do not apply any extinction correction for the g-band absolute magnitude and note again that our metallicity results have a “ceiling effect” at 12+log(O/H)T e ∼ 8.9. Our samples share a similar result with Guseva et al. (2009) if we assume 12+log(O/ H)T e > 8.15. As discussed above in Section 6.3, our R23-method metallicity estimation 34.

(44) Figure 13: Stellar Mass-Metallicity Relation(MZR). Symbols and colors are same as Fig. 9. Open symbols correspond to N2-method results and filled symbols represent the R23method results. Black, green, orange and red solid lines show the mass-metallicity relation at z ∼ 0.07, 0.7, 2.2 and 3.5 (Maiolino et al. 2013). Our loose-, middle- and tight- photometric criteria show as top, middle and bottom panels. Left panel: The metallicity of some samples are derived by R23-method which assume that the 12+log(O/H)T e < 7.95. Right panel: Same as left panel but using R23-method that taking assumption of 12+log(O/H)T e > 8.15. The right panel can be considered as the upper limit of metallicity for our samples. Because of the double possible metallicity values, we cannot judge if there is a significant difference in metallicity among the loose-, middle- and tight-criterion samples.. 35.

(45) Figure 14: Luminosity-Metallicity Relation(LZR). Symbols and colors are same as Fig. 13. Dotted and dashed lines represent the relation of emission line galaxies studied by Guseva et al. (2009) and luminous compact galaxies (LCGs) samples from Itozov et al. (2011a) for tight-criterion selected samples is more likely single-value at 12+log(O/H)T e ∼ 8.0. The LZR for our tight-criterion selected samples could be more consistent with the results from luminous compact galaxies (LCGs) samples (dashed line in Fig. 14) which share a similar properties with metal-poor star-forming galaxies at z ≤ 1.0 (Izotov et al. 2011a). However, same as the result with MZR, we cannot judge if there is a significant difference in metallicity among the loose-, middle- and tight-criterion samples.. 6.5 Redshift versus SFR Indicators Luminosity We represent the relation of redshift and luminosity of SFR indicators for the entire emission line galaxies samples (Fig. 15). The loose-, middle- and tight-criterion are shown in panels (a), (b) and (c). The filled and open symbols correspond to the [OII]λ3727Åand 36.

(46) Hα luminosity, respectively. We also include the SDSS-peas galaxies and NTT-peas galaxies samples which are in color blue and light blue. The SDSS-peas samples have both OII and Hα detections. Thus, we show both luminosity of [OII]λ3727Åand Hα which display a comparable order of luminosity in Fig. 15. Interestingly, our tight-criterion is not effective to select the objects which are high in luminosity of SFR indicators (Fig 15c). It is because our tight-criterion selected samples are high in SSFR .i.e high SFR at a given stellar mass. On the other hand, the SDSSpeas samples are also selected by photometric selection criterion. However, they lead to a different result that show extremely high in luminosity of SFR indicators. This feature could be caused by the SDSS-peas samples are more extreme case compared with our tight-criterion selected samples in the color-color diagram (green star symbols in Figs. 3a, 4a, 5a). However, we do not have many spectroscopic observations in our tight-criterion. This might lead us to the bias that we do not filter out the most extreme “green pea”- like samples from our tight-criterion in color-color diagrams. The luminosities of SFR indicators show an upper limit ∼1043 ergs−1 , which is independent of redshift (dotted line in Fig. 15). This feature was also described by Izotov et al. (2011a) except that they are looking at Hβ luminosity and the upper limit of L(Hβ) ∼ 2.5×1042 ergs−1 . If we adopt the theoretical ratio for Case B, L(Hβ)=L(Hα)/2.86, the upper limit of L(Hα) ∼ 7×1042 ergs−1 is consistent with our results. The sample from Izotov et al. (2011a) have lower redshifts (up to z ∼ 0.6). Our results, which contain higher redshift objects, imply that this feature is remarkable not redshift dependent. Izotov et al. (2011a) claims that this is probably a negative feedback of mechanism due to the intense UV radiation in these high star-formation samples. However, our entire samples show that this might be just a feature commanly seen in any readshift versus absolute magnitude diagram. There is a target at z∼1.35, which shows a [OII] luminosity above the limitation, could be very interesting. However, the spectroscopic observation of this target shows that the ˚ ). The broad emission line imply that this target [OII] emission line is quite broad (∼ 35 A might have the AGNs containimation which can barely selected out by BPT-diagram in. 37.

(47) Figure 15: Redshift versus SFR indicators luminosity diagram of our entire emission line galaxies samples. The loose-, middle- and tight-criterion are shown in panels (a), (b) and (c). The OII and Hα luminosity correspond to the filled and open symbols, respectively. We also include the SDSS-peas galaxies and NTT-peas galaxies samples which are in color blue and light blue. The SDSS-peas samples have both OII and Hα detections. Thus, we show both luminosity of OII and Hα which display a comparable order of luminosity. The luminosity of SFR indicators show a upper limit ∼1043 ergs−1 that independent with redshift (dotted line). this such of high redshift.. 38.

(48) Chapter 7 Summary We developed a method based on broad-band photometry to isolate objects similar to “green pea” galaxies (Cardamone et al., 2009) and Luminous Compact Galaxies (LCGs) (Itozov et al. 2010a) in some various redshift ranges up to z∼1.5. We extend our method to several deep photometric archive datasets which are UKIRT infrared deep sky survey - Ultra Deep Survey (UKIDSS-UDS), Canada-France-Hawaii Telescope Legacy Survey (CFHTLS) and COSMOS-UltraVISTA survey. For convenience, all the magnitudes in optical and NIR filter sets in each catalog are converted to those of the SDSS(u, g, r, i, z) and UltraVISTA(J, H, Ks) filters. To define our selection criteria, we use 80 low-redshift SDSS peas spectra (z∼0.3, Cardamone et al., 2009) and 8 higher-redshift ESO-NTT peas spectra (z∼0.6, Bamford et al., 2014) as spectrum templates. Convolving these spectrum templates with broad-band photometric system enables us to define the peas selection criteria in color-color diagrams. Thanks to the deep photometric datasets available (UDS-DR10, CFHTLS+Palomar and COSMOS+UltraVISTA), we can extend our broad-band method to select the r-, i-, z-, Jpeas which represent the “green pea” galaxies at z∼0.25, z∼0.55, z∼0.85, z∼1.5 for all the photometric catalogs. We also combine the spectroscopic data for each of our catalogs. Then, we select a very loose photometric criterion by redshift distribution of the spectroscopic data. We further divide our spectroscopic data into two more smaller parts: middle- and tight-criterion which are more consistent with the distribution of “green pea” galaxies in color-color 39.

(49) diagrams. From the emission line information of our spectroscopic sample, we obtain the following results: 1. The global properties of our tight-criterion selected objects are low in stellar mass (M∗ ∼109 M⊙ ) and low metallicity (12+log(O/H)∼8.0) compared with normal galaxies. However, they display high in star formation rate compared with their given stellar mass (SSFR∼10−9 yr−1 ). 2. The results of K-S test for SSFR represent that our tight-criterion selected objects are significantly different from loose-criterion selected objects. This indicates that our tight-criterion is efficient to select the high SFR objects in their given stellar mass. 3. Because the temperature sensitive line [OIII]λ4363Åis too weak to be observed, we ultilize the emprical methods (N2-method and R23-method) to derive the metallicity for our samples. The results of mass-metallicity relation (MZR) and luminosity-metallicity relation (LZR) show that our samples are consistent with others’ work. However, the MZR and LZR results show that our tight-criterion selected objects are not significantly different with middle- and loose-criterion selected objects. This analysis is, however, limited by the double-value solutions for the metallicity derived by the R23-method. 4. If we only look at the metallicity which is derived by R23-method, our tight-criterion selected objects are almost around 12+log(O/H)T e ∼ 8.0. This could imply that our tightcriterion selected objects are more likely single-value at 12+log(O/H)T e ∼ 8.0. The MZR for our tight-criterion selected samples might be similar to the high redshift objects. Beside that, the LZR for our tight-criterion selected samples could be more consistent with the luminous compact galaxies (LCGs) samples which share similar properties with metalpoor star-forming galaxies at z ≤ 1.0 (Itozov et al. 2011a). 5. The luminosities of SFR indicators show an upper limit ∼1043 ergs−1 , which is independent of redshift (dotted line in Fig. 15). This feature was also described by Izotov et al. (2011a) except that they are looking at Hβ luminosity and the upper limit of L(Hβ) ∼ 2.5×1042 ergs−1 . If we adopt the theoretical ratio for Case B, L(Hβ)=L(Hα)/2.86, the upper limit of L(Hα) ∼ 7×1042 ergs−1 is consistent with our results. The sample from. 40.

(50) Izotov et al. (2011a) have lower redshifts (up to z ∼ 0.6). Our results, which contain higher redshift objects, imply that this feature is remarkable not redshift dependent. Izotov et al. (2011a) claims that this is probably a negative feedback of mechanism due to the intense UV radiation in these high star-formation samples. However, our entire samples show that this might be just a feature commanly seen in any readshift versus absolute magnitude diagram.. 41.

(51) Appendix A Conversions between Photometric Systems For the convinence, all the optical filters in each of the catalogs are transfered into the SDSS base filters. We adopt the following equations to do the conversion between UDS and COSMOS with the SDSS filters. g SDSS = V + 0.72 (B - V) - 0.13 (similar with the equation from Jester et al. 2005) r SDSS = r’ + 0.035 (r’ - i’ -0.53) - 0.15 (SDSS website) i SDSS = i’ + 0.041 (r’ - i’ - 0.21) (SDSS website) z SDSS = z’ - 0.03 (i’ - z’ - 0.09) (SDSS website) The transformation for the UDS Rc filter to r’ is given by: r’ = V - 0.84 (V - Rc U DS ) + 0.13 (Fukugita et al. 1996) In the case of CFHTLS, the terms between MegaCam and SDSS filter can be described by following equations: g SDSS = g M ega + 0.195 (g M ega - r M ega ) r SDSS = r M ega + 0.011 (g M ega - r M ega ) i SDSS = i M ega + 0.079 (r M ega - i M ega ) z SDSS = z M ega - 0.099 (i M ega - z M ega ) Those equation are quote from the CFHTLS website and all the transformations are in AB magnitude.. 42.

(52) On the other hand, all the NIR filters data are transfered into UltraVISTA filter base observation. Because of the Palomar filter instruments are resembling the UltraVISTA filter base, we only achieve the filter transformation from UDS WFCAM to UltraVISTA VIRCAM. J2M ASS = JU DS - 0.01 - 0.03 (JU DS - HU DS ) + 0.1 H2M ASS = HU DS + 0.01 + 0.065 (HU DS - KU DS ) Ks2M ASS = KU DS - 0.072 (HU DS - KU DS ) Remind here that those equations (Hewett et al., 2006) are base on Vega magnitude. The offsets within Vega and AB magnitude are display as: JAB = JV ega + 0.938 HAB = HV ega + 1.379 KsAB = KsV ega + 1.9 Although the 2MASS filter sets are tied to UltraVISTA photometry systems, they still need to do slightly calibration between them by (From the VISTA website): JU V IST A =J2M ASS -0.077 (J2M ASS - H2M ASS ) + 0.1 HU V IST A =H2M ASS +0.032 (J2M ASS - H2M ASS ) + 0.1 KsU V IST A =Ks2M ASS +0.01(J2M ASS - Ks2M ASS ). 43.

(53) References Baldwin, J. A et al. 1981, PASP, 93, 5 Bamford et al. in prep Bigiel et al. 2008, AJ, 136, 2846 Blanton et al. 2007, AJ, 133, 734B Bundy et al. 2006, ApJ, 651, 120 Cardamone et al. 2009, MNRAS, 399, 1191 Cowie et. al., 1996, AJ, 112, 839 Daddi et al. 2010a, ApJ, 713, 686 Daddi et al. 2010b, ApJ, 714, L118 Fukugita et al. 1996, AJ, 111, 1748 Guseva et al. 2009, A&A, 505, 63 Hewett et al. 2006, MNRAS, 367,454 Izotov et al. 2011a, ApJ, 728, 161 Izotov et al. 2013, A&A, 561, 30 Ilbert et al. 2013, AA, 556, 55 Jester et al. 2005, AJ, 130, 873 Kennicutt et al. 1998, ApJ, 498, 541 Kewley et al. 2001, MNRAS, 372, 961 Liang et al. 2006, ApJ, 652, 257 Lilly et al. 2007, ApJS, 172, 70 Newman et al. 2013, ApJS, 208, 5N Maiolino et al. 2008, A&A, 488, 463. 44.

(54) McCracken, H. J., et al. 2012, A&A, 544, 156 Pagel B.E.J., et al. 1979, MNRAS, 189, 95 Pettini et al. 2004, MNRAS, 348, 59 Pilyugin et al. 2000, A&A, 362, 325. 45.

(55)