In [12]:
import numpy as np
from pylab import *
%matplotlib inline
import warnings
warnings.filterwarnings('ignore')
We need to decide what our stellar mass limit will be. We need to think about this because at a given absolute magnitude limit, the lowest SFR objects will have the highest $M_*/L$ and hence the highest $M_*$. Converseley at a fixed $M_*$ cut, the lowest SFR objects will be the faintest. So, if we make our $M_*$ cut too low, objects with low SFR will fall below our detection limit.
Could cause a bias such that we miss the most suppressed SF-galaxies, or those with the smallest 24um disks?
In [13]:
run ~/github/LCS/python/Python3/LCS_MSpaper.py
This shows how all the different selection cuts manifest themselves in the SFR-$M_*$ plane. In all plots, red are the objects removed by the flag.
The first plot is for all the galaxies and the second and third are separated by core and exterior.
The horizontal green line is the SFR corresponding to our LIR limit. I took the 0.086 from Elbaz and divided it by 1.74 to convert to Chabrier (Salim+16).
The blue line is my fit to the MS (see below for details.)
In [9]:
g.plotSFRStellarmasssel()
In [11]:
g.plotSFRStellarmasssel(subsample='core')
In [12]:
g.plotSFRStellarmasssel('exterior')
This shows the SFR-$M_*$ distribution for all galaxies and those that pass the final selection. It seems that making a cut at log($M_*$)>9.5 excludes galaxies at the bottom of the main sequence. The cut below, at log($M_*)>9.7$ seems to work.
The solid lines are for SFR(MS), the dashed for SFR(MS)/5. Black is for Elbaz+11, Salmon is for Salim+07 for their pure SF sample, blue is our fit to the non-AGN galaxies above our LIR limit with log(Mstar)>9.5 and with SFR>SFR_MS(Elbaz; Mstar)/6.
This seems to show that there isn't any different selection between core and cluster. However, it does seem that the Elbaz and Salim lines lie above ours. The difference is much in excess of the factor of 1.58 that you would expect if there were some IMF mismatch.
In [4]:
g.plotSFRStellarmassall()
g.plotelbaz()
g.plotsalim07()
In [5]:
g.plotSFRStellarmassallenv()
On the left I show all the galaxies in our sample and on the right the running median. Both environments have similar median SFRs but different sizes. The solid blue line is a fit to the non-AGN galaxies above our LIR limit with log(Mstar)>9.5 and with SFR>SFR_MS(Elbaz; Mstar)/10.
In [13]:
g.plotSFRStellarmass_sizebin()
g.plotSFRStellarmass_sizebin(btcutflag=False)
galaxies in the core have smaller R24/Rd than external galaxies at most stellar masses
In [16]:
g.plotSFRStellarmass_musfrbin()
galaxies in the core have higher muSFR than external galaxies at lower stellar masses. However, the external galaxies have higher muSFR at higher masses. This result doesn't change if I color code by log(median(musfr)) or median(log(musfr)). Also, the result does not qualitatively change if I remove the B/T cut.
In [17]:
g.musfr_size()
muSFR is computed as 0.5 SFR / (pi R242). The dashed line is y = log10(1/x2) and so shows the expected correlation. It is not clear that there is any residual correlation.
In [22]:
g.sizediff_musfrdiff_mass()
This shows that the sizes are systematically lower in the core. It also shows that the SFR surface densities show no clear trend with mass of the ratio between the two environments. Also, the result does not qualitatively change if I remove the B/T cut. Finally, the results don't change if I do them in slices around the main sequence.
In [6]:
g.sfr_offset()
g.sfr_offset(btcutflag=False)
Although there is a tail to lower SFR in the core, we can't reject the hypothesis that these are drawn from the same distribution. Below I'm showing some versions of this in different mass slices.
In [20]:
g.sfr_offset(logmassmin=9.7,logmassmax=10.3)
In [21]:
g.sfr_offset(logmassmin=10.,logmassmax=10.5)
In [22]:
g.sfr_offset(logmassmin=10.3,logmassmax=10.8)
Doesn't look like the differences become significant with different mass bins. At least not without fiddling a lot with the binning.
In [23]:
g.plotmusfr_optdisksize()
At a fixed $R_e(r)$, galaxies with lower $\Sigma_{SFR}$ also have lower $\Sigma_{\star}$. Clusters are missing a small number galaxies at high $R_e(r)$ with large $\Sigma_{SFR}$ but it's not clear if it's significant. My guess is that's what driving the poor correlation in the external sample. At fixed $R_e(r)$ it also appears that the cluster could have $\Sigma_\star$ than the external sample but it's not clear. $Sigma_\star$ is measured using the single Sersic fits to the galaxy from the NSA. We might want to do these using GIM2D to be consistent. Or compute the $R_e$ analytically from the Bulge-disk fits, again to be consistent.
The horizontal dashed line is for reference only
In [24]:
g.plotmusfr_mustar()
This might just be telling us that smaller stellar disks also have smaller SF disks. But interestingly, it's not just size but also surface density.
In [25]:
g.musfr_mustar_ks()
This shows the distribution of $\Sigma_{SFR}$ and $\Sigma_\star$. While the probability that these are different distributions is high with this mass binning as I show below, the significance of the difference goes away with a slightly different mass binning. So I would say that they are identical distributions
In [26]:
g.musfr_mustar_ks(logmassmin=9.6, logmassmax=10.8,dlogmass=0.3)
In [27]:
g.matchsamp_mass()
This plot shows mass-matched samples constructed by choosing all external galaxies within +/-0.3 dex in mass. I omit core galaxies with no field galaxies in this mass range. Each point shows how one of the core galaxies differs from the value for the median of the matched sample. The red line shows the intrinsic correlation you'd expect in the right-hand axis, given that $\Sigma_{SFR}$ is inversely proportional to $R_{24}^2$. This assumes that the optical disks are the same size and that the SFRs for the two samples are the same.
The blue line shows a simple fit to the black points assuming uniform errors. It appears that there is no residual correlation aside from what is expected. I wonder what would happen if I included errors or bootstrapped the line.
In [28]:
g.plot_n24diff_mipssizediff()
g.plot_n24diff_mipssizediff(btcutflag=False)
This is now constructed using samples matched in stellar mass and optical disk size. The left panel is color-coded by the difference in SFR. From van der Wel et al. 2012 Fig. 7 it looks like bigger sizes will result in slightly larger sersic n. This is the opposite trend from what we see in the left panel, which indicates that larger sizes yield lower sersic n. It also looks like galaxies with lower $n$ also have lower SF. Also from van der Wel et al. 2012, it appears that making the sersic $n$ larger makes the integrated profile brighter. That would mean that increased n could result in systematically higher SFR, which could drive some of the correlation of $\Delta \Sigma_{SFR}$ with $\Delta n_{24}$
In the right panel, the scatter is lower than in the matchsampmass plot, indicating that the scatter may have been driven by the scatter in optical disk size. The green line is the median $\Delta log \Sigma{SFR}$ There may be a weak trend but I need to do a better job with the errors to confirm this. The red line is the expected correlation and the blue line is the fit (without data uncertainties) to the data
In the plot below, I highlight the sources with $\Delta log(n_{24})>0.45$ and $\Delta log(SFR)<-0.2$. These are the red/orange points at the top of the plot. These are the x's in the figure below
In [11]:
g.plotSFRStellarmass_matchsamp()
g.plotSFRStellarmass_matchsamp(btcutflag=False)
In [10]:
g.sfr_offset_matchsamp(btcutflag=False)
This shows the distribution of SF w.r.t. the main sequence for different samples of galaxies. Top: all the galaxies in the SF sample. Second: all the galaxies in the morphological sample. Third: the galaxies in the morphological sample with normal 24um sizes compared to a set of matched external galaxies. Bottom: the galaxies in the morphological sample with small 24um sizes compared to a set of matched external galaxies. All of the samples are consistent except for the top and second panels, which are marginally consistent. This doesn't really depend on where I draw the boundary between small and normal.
In [11]:
g.sersicint()
This plot shows how the luminosity of a sersic profile depends on the radius at which I cut the profile off. This is for different sersic indices and all curves are normalized to to have the same central intensity. The total luminosity is computed at R<7Re.
In [14]:
g.r24_re_msdist()
g.r24_re_msdist(btcutflag=False)