A concurrent SFRD and BHARD in COSMOS: Blending disjointed samples & Testing AGN SED Decomposition methods

Context: The possible co-evolution between galaxies and their central supermassive black holes is supported by the similarity in shape between the Star Formation Rate Density (SFRD) and Black Hole Accretion Rate Density (BHARD) of the universe. Both trends (upper right) peak around z~2 and decline to present day, however, due to selection effects they do not trace the same galaxies and thus represent a similarity in the global evolution of these two quantities, rather than local. To trace the local causal connection, if any, requires constructing both trends for the same sample in a self-consistent manner. 

Star Formation

Rate Density

Black Hole Accretion 

Rate Density

Screen Shot 2019-02-21 at 9.13.57 PM.png

What's new?: We will construct the SFRD and BHARD for the exact same sample of galaxies for the first time, also exploring a full range of AGN template parameter space. Previous BHARDs use ~35% of IR-selected samples to ensure strong AGN, however, nearly ~50-70% of IR samples can contain sources with measurable AGN contribution. This Herschel-250um-selected sample from COSMOS includes ~7,000 galaxies and we perform a series of AGN SED Decomposition tests with the MIT Supercloud Supercomputer to map systematics associated with decomposition of weak AGN, allowing for a statistically larger sample that does not conservatively omit these potentially bolometrically significant sources. The tests also probe how sparse broadband MIR data (a limitation of large high-z samples) impacts SED decomposition, along with AGN SED model parameters, paying close attention to the presence/depth of the Si 9.7um absorption feature in AGN SED. 

Method: We use a test-sample of~100 24um-Selected sources with Spitzer/IRS MIR spectroscopy, 0.4 <z <2.7, and ranging in MIR AGN strength from Kirkpatrick+2012. Allowing the full range of AGN SED theoretical models from Fritz+2006 and Feltre+2012, commonly used in the literature, we study broadband decomposition results (SED3FIT; Berta+2013)  and compare best-fit AGN fractions and parameters with those derived from MIR spectroscopy in Kirkpatrick+2012 to identify systematics. We see how these fits perform with a limited set of AGN templates and a wider set, also identifying where in the rest-frame MIR broadband coverage is most sensitive/crucial to constraining results.  

Results: Paper I details SED fitting testing and is in prep, while the remainder of this thesis work (early 2021, Paper II) will compute both trends. Paper I  preliminary results are as followed:

1. Higher optical depth AGN models (with extended FIR emission) offer better fits to AGN sources, suggesting high amounts of AGN dust-heating from either the nuclear region or the galaxy outskirts

2. AGN MIR fractional contribution can be underestimated by ~30% via SED fitting if there is no broadband constraint in the rest-frame ~5-12um region. 

3. Absence of a broadband constraint in the narrow rest-frame ~5-8um region can falsely overestimate MIR AGN fraction by ~20-30%.

4. Weaker AGN can be well-fit by both high optical depth AGN models (strong Si absorption) and weak Type-1 AGN, resulting in AGN luminosity difference of ~2 dex. This demonstrates the sensitivity of allowed AGN models in fitting and supports JWST MIRI follow-up to understand the nature of these sources.

Screen Shot 2019-02-21 at 9.31.58 PM.png
Screen Shot 2019-02-21 at 9.31.54 PM.png

Star Clusters in Dwarf Galaxies
Universality of the IMF

Context: The predominance of low-mass clusters in dwarf galaxies necessitates stochastic sampling of the IMF to accurately predict their mass and subsequently, age.  A randomly sampled IMF introduces the probability that a low mass cluster might host a massive star. The presence of one or more massive stars will concede notable fluctuations in the ionizing photon luminosity from cluster to cluster. Understanding this scatter and capturing its ramifications while deriving cluster properties is extremely important for low SFR environments where stochastic effects can mimic that of a varying IMF.

What's new?: Age, mass and extinction of these unresolved star clusters are obtained using the Bayesian SED fitting routine SLUG that implements stochastic sampling of the stellar Initial Mass Function (IMF) utilizing U to I SEDs with Hα emission from each cluster. The use of stochastic sampling for determination of cluster properties is particularly imperative for dwarf galaxies due to their low star formation rate densities (ΣSFRs) and consequentially low mass clusters (< 10^3M⊙). An example SLUG PDF is shown below.

Method: With ages and masses determined with SLUG, we compare observations of Hα luminosity normalized by cluster mass with SLUG predictions from three varying IMF forms: (1) Universal Chabrier IMF, (2) Integrated Galactic Initial Mass Function (IGIMF) of Weidner & Kroupa (2006), and (3): IGIMF of Weidner et al. (2010). The IGIMF of Weidner et al. (2010) truncates the IMF based on a strict upper-mass limit for the most massive star in the cluster, mmax, that is determined by the total cluster mass, Mcl, resulting in systematically less massive stars. The Weidner & Kroupa (2006) formulation of the IGIMF is similar to Weidner et al. (2010) but instead makes it less likely, rather than impossible, for a cluster to form stars more massive than the mmax limit.

Results: 

1. We detect a total of 191 star clusters in NGC 2366 and IC 2574, with 85 clusters in NGC 2366 (60 with Hα and 25 without Hα) and 106 clusters in IC 2574 (64 with Hα and 42 without Hα). UBVHα I SEDs for each cluster are constructed using a combination of multi-wavelength HST ACS and ground-based images
 

2. We add new data points to the L(Hα)/Mcl −Mcl relation, previously investigated by Calzetti et al. (2010) and Andrews et al. (2013, 2014) for M51a, NGC 4214, and M83. Our data for NGC 2366 and IC 2574 extends to lower cluster mass bins and is consistent with the scatter in their values.

3. Since the L(Hα)/Mcl −Mcl relation is age-sensitive, we com- pare the simulations categorized by two age cuts: (1) Simulations match age distributions of observations via Monte Carlo sampling, and (2) Simulations include all possible ages 1-8 Myrs. After comparing possible combinations of distrbutions and performing KS tests, we find that, of the three IMF models, the WK06 IGIMF resembles the observed distribution most closely considering the full possible range of observed distributions with completeness corrections.

Prestellar Cores in MonR2 Molecular Cloud

with the Large Millimeter Telescope