~~~~~~~~~ Notes from thesis defense on 4 August 2009 Presentation available at http://www.ucolick.org/~kcooksey/MLSS.html ~~~~~~~~~ Probing the Chemical Composition of the z < 1 Intergalactic Medium with Observations and Simulations Thank you for attending the public portion of my defense. I will be discussing the chemical composition of the low-redshift (z < 1) IGM. I would like to introduce my committee: my advisor, Xavier; Hsiao-Wen Chen from the University of Chicago and co-author on my two papers; and Anthony from the physics department, with whom I'm finally in a position to work. (We talked about using simulations and "observing" with synthetic spectra back in my first year, and only recently have I been in a position to do it.) Questions are welcome. I trust the committee to decide when they should ask questions during the public portion and when they should save it for later. ~~~~~~~~~ Outline I will be motivating why you should care about this research first (obviously). Then I'll describe the technique briefly. The majority of the talk will be about the observations and comparing "simulated observations" to observations. I'll quickly list the future work before I summarize. The plot is a teaser, indicating that I'll be measuring the mass density of triply-ionized carbon and comparing it to the copious work done at z > 1.5. ~~~~~~~~~ SCIENCE DRIVERS ~~~~~~~~~ ~~~~~~~~~ Context Let me frame the broad context. These are the fundamental questions I have. They are important questions because they affect e.g., star formation, galaxy colors and morphologies, etc. How many metals are produced in the Universe? This is a question about star formation and anything/everything that affects it. Where are they? We, who are made of many metals, are not in stars, where the metals are produced. Metals are observed in many places in the Universe. How did they get there? We observe metals not in stars. We also observe metals being released from stars in a variety of ways. I will be observing the metals that are not in galaxies (though they may be close, hint, hint), and I will turn to simulations to see which physics affect simulated observations (to better understand the processes that may determine what I observe). ~~~~~~~~~ Cosmic Chemical Enrichment Cycle The whole time that stars are forming and dying you have chemical enrichment, and it's a cycle. The stars produce the metals. Those metals will end up in the ISM, halo, and/or canonical IGM. They get there through some outflow mechanism. Wherever they may go, they could end up in a star again. Some of the "more escaped" material may rain back into a galaxy. ~~~~~~~~~ "Feedback" Let me give my working definition of feedback. Feedback is stars and/or AGN moving material (gas, metals, etc) and energy. We observe these processes as winds, jets, bubbles, etc in galaxies. The movement of material and energy affects the e.g., star formation rate (if there's less gas or it's higher temperature gas, star formation will be suppressed). Also, feedback may change the physical condition of the gas (temperature, density) which changes what ions are observed. The distribution of the gas is affected. For example, here are 100 kpc/h boxes of the same z=2 galaxy in a suite of simulations with different physics. Feedback affects the immediate vicinity of the source of the feedback as well as things farther away (and how far is "far" depends on many factors). And feedback happens over all time. ~~~~~~~~~ QUASAR ABSORPTION-LINE SPECTROSCOPY ~~~~~~~~~ ~~~~~~~~~ Quasar Absorption-Line Spectroscopy I'll be using quasar absorption-line spectroscopy to study the metals in the IGM. Here is a classic diagram of how quasar absorption-line spectroscopy works. Light from the background quasar is absorbed by the intervening gas at wavelengths characteristic of the elements in the gas. The light passes through the quasar's local environment (intrinsic), the cosmic web (intergalactic), and our Galaxy's gas. The most direct observables for quasar absorption lines are: the number of lines of interest, their redshift (after you identify them); their rest equivalent widths (the strength of the absorption); and the column density (which we measure by the apparent optical depth method, so it depends on the amount of absorbed flux). Column density is a physical property of the gas, whereas equivalent width is the observed quantity. So most of this talk deals with column densities. ~~~~~~~~~ QAL Spectroscopy in Action Here's a movie demonstrating how the light traveling from the background quasar is absorbed by the intervening material in a fashion characteristic of the material (elements, distribution, etc). High-redshift spectra have a lot of lines. At z < 1, I don't have to suffer that much. But notice that red-ward of the quasar Lya emission line, the density of lines is less. There are only metal lines there. ~~~~~~~~~ Metal Absorption Lines Affected by... (And I'm mostly referring to the number of lines observed and their column density here.) More metals and more of the particular metal of interest can increase the number and/or strength of the transitions observed. The ionizing background changes the favored transitions. The physical distribution will change the number of lines and their strengths. ~~~~~~~~~ Ultraviolet Background and Ionization Balance Here's a quick plot of how the UV ionizing background affects the strength of transitions in a optically thin, plane-parallel slab of gas with solar relative abundances and metallicity one-tenth solar. As the background intensity increases, the carbon becomes more highly ionized. ~~~~~~~~~ Physical Distribution The number of lines observed changes from high redshift to low. But that doesn't mean that the number of lines per redshift changes. Even more important is to look at the absorber line density dN/dX, which is the product of the co-moving number density of the lines and their physical cross section (also co-moving). Evolution (or no evolution) in dN/dX has physical implications about the distribution of the gas through time. ~~~~~~~~~ OBSERVATIONS (paper title and co-authors) I have studied other metals but I'm focusing on CIV here. ~~~~~~~~~ ~~~~~~~~~ Luminosity Function Analogy One goal of my thesis was to measure the column density frequency distribution of whatever ion was of interest (CIV actually). The frequency distribution is similar to the luminosity function of galaxy surveys. Since there are many galaxy people in the audience, I just wanted to run through the similarities between these two quantities. The luminosity function is the number of galaxies per luminosity bin per unit volume surveyed where such galaxies can be observed. I'm using magnitude and luminosity interchangeably here. Once you have the luminosity function, you can integrate it to measure the e.g., total number of galaxies per unit volume or the total luminosity density. Measuring the volume surveyed is crucial. It's what dictates the faint-end slope. The same is true for the frequency distribution. A lot of time is spent (has been spent) on estimating the co-moving pathlength dX sensitive to the observed absorbers (their column densities). ~~~~~~~~~ Archival Spectra We culled the HST archive for targets with STIS and/or GHRS (UV) spectra. Of the almost 70 targets that had such data, only 49 had spectra with the right wavelength coverage (CIV is at 1548, 1550 Ang), resolution, and signal-to-noise. We incorporated any FUSE observations of the same targets for coverage of shorter wavelength transitions. The STIS spectra made up the bulk of the coverage. The echelle gratings were really the movers and shakers of the survey. The E140M grating covers CIV up to z ~ 0.1, and there's a lot of E140M observations. The E230M grating covers CIV from z ~ 0.4 to 1, depending on the tilt (it can cover down to z = 0). The other STIS gratings and the GHRS spectra make up the rest. Across the top is marked the redshifts of the final CIV sample. ~~~~~~~~~ Automatic, Blind CIV Survey We conducted a (mostly) automated, blind survey. First we automatically detect absorption lines. Then we have the computer pair up all lines that have roughly the characteristic separation of the CIV doublet. In a similar fashion, the computer tacks on other common transitions (such as HI Lyman lines). In a velocity plots, the various transitions associated with one system are stacked in velocity space, determined by the rest wavelength of the transition and the redshift of the system, in this case, the redshift of the 1548 line. They show how well aligned the transitions are and how the profiles compare. We assign flags to the candidate CIV systems based on ideal characteristics of actual CIV systems, such as how similar their line profiles are or if they have other lines associated with them. Then we visually evaluated the candidates, voting for the ones we were confident were actual CIV systems, and adjusting a few details (e.g., boundaries). ~~~~~~~~~ Final CIV Sample In the end, we had two groups. G = 1 is the group of definite CIV absorbers, and the G = 2 group is the likely CIV systems. We'll only analyze systems with both doublet lines detected at greater than 3-sigma in equivalent width. (Other studies make other cuts.) This leaves us with 38 G = 1 doublets. And I will only talk about the G = 1 sample; adding the G = 2 sample does not change the results. For some analyses, we divide the G = 1 group at roughly the median redshift to see what we can with regards to evolution. There's a slew of other checks we made that of course took a lot of time but did not merit more than a mention. HI Lyman lines are the biggest contaminate to our survey, which we did _not_ restrict to outside the Lya forest, so we had to characterize how that affects our results. (Not much.) We used what published results there were as checks to our methods. We made our own line lists. We did a special check that the CIV doublets weren't actually OI 1302, SiII 1304 at another redshift; this pair has about the same wavelength separation as the CIV doublets. Since our automatic blind search depends relies solely on the separation, this was a smart check (thanks to Edward Jenkins for telling us to do it). ~~~~~~~~~ Survey Sensitivity As I mentioned in the luminosity function analogy, estimating the co-moving pathlength (equivalent to volume) surveyed and sensitive to the absorbers of different strengths is crucial. We Monte-Carlo-ed absorbers with a range of column densities, widths (Doppler parameters b), redshifts, and number of components into the actual spectra (that had been "cleaned" so that it was just continuum with the real noise properties). Then we measured the column density at which 95% of the input _doublets_ were recovered automatically. In the plot, we have the column density limit as a function of the redshift of the input doublets. The spikes are saturated pixels in the Lya forest (which contain no information). The shaded lines are the Galactic lines (counted twice). On the right is the total pathlength function for this one sightline. On the bottom is the column density histogram for the input and recovered lines. ~~~~~~~~~ Survey Sensitivity Repeating this procedure for all sightlines (all their spectra) we generate dX as functions of column density and equivalent width. We also have dX for the redshift bins. ~~~~~~~~~ N(C+3) Frequency Distribution Now we can compute the column density frequency distribution f(N). It looks like a power-law, so we fit that through a maximum likelihood method that includes the number of saturated doublets as a constraint, because we only have lower limits on their column densities. For the full G = 1 sample, the exponent is -1.5+/-0.2, that includes 36 CIV doublets. ~~~~~~~~~ f(N(C+3)) by Redshift Dividing the G = 1 sample at the median redshift and fitting power laws shows us that there isn't significant difference between the two sub-samples. So there isn't significant evolution, even compared to z > 1.5 studies. In case you wonder whether the different, dominating STIS gratings would produce different results (indicating some systematic effect of the e.g. different spectral resolution), it's not so. ~~~~~~~~~ Absorber Line Density I mentioned the absorber line density earlier. We actually measure it by weighting the absorbers above the limit of interest with dX. The 0th moment of the frequency distribution (or the integral over column density) is the fit. Just a reminder, because it becomes important later, this quantity dN/dX is the result of the distribution of the absorbers... the product of their co-moving number density and cross section. ~~~~~~~~~ dN/dX: Evolution? We look at the (redshift) evolution of the absorber line density by comparing to the integrated value from Songaila (2001). People have not published the information for me to do better than this. The background points are the summed values for absorbers with log N(C+3) >~ 12 (actually not properly corrected). The increase in dN/dX from z = 3.2 to z = 0.6 is detected with statistical significance. But in light of the fact that 8 Gyr separate these measurements, it's intriguing that the physical distribution (co-moving number density and cross section) have conspired to change by less than a factor of 2. ~~~~~~~~~ C+3 Mass Density I mentioned that measuring the C+3 mass density at z < 1 was a motivator for this project. Here is the definition of Om(C+3), the C+3 mass density relative to the critical density of the Universe. Again, integrating the frequency distribution to get the mass density is very similar to taking the first moment of the luminosity function to measure the luminosity density of the Universe. All the other studies to which we compare use the sum of the column densities to measure the mass density. They typically include all absorbers they detect, which is log N(C+3) = 13 to about 15. We can't do that because of saturated absorbers and how we measure the column. We would only have a lower limit to the 13 <= log N(C+3) <= 15 Om(C+3). So we use the integral of our best-fit power law. See here, for our alpha > -2 power law slopes, the mass density is infinite and dominated by the stronger absorbers. ~~~~~~~~~ C+3 Mass Density: Over Time Om(C+3) appears to have increased. But by how much and is it significant? ~~~~~~~~~ C+3 Mass Density: Evolution? The error-weighted average of the 1.5 < z < 5 studies is (2.2+/-0.18)x10^-8. Our measured z ~ 0.6 Om(C+3) is 2.8+/-0.7 time greater. So that's significant. ~~~~~~~~~ C+3 Mass Density: Evolution? We do the next logical thing and fit a line to the data as a function of time. This is no more than a toy model. The slope is unconstrained for the 1.5 < z < 5 data, so it wasn't known whether they were evolving or not because the data allowed for anything. But including our data points, we have a 2-sigma detection of the slope. The low-redshift data adds the evidence to show that Om(C+3) is evolving, even from z = 5 to 1.5. ~~~~~~~~~ Observations Summary We have shown that the co-moving number density and the cross section of absorbers have conspired to make dN/dX change by less than a factor of two in 8 Gyr. The 13 <= log N(C+3) <= 15 C+3 mass density has increased almost threefold in the same time. Therefore, the typical column density of the absorbers at z < 1 is higher than at z = 1.5 to 5. This suggests that CIV absorbers are tracing higher density gas at z < 1. We can even go so far to say that they're tracing circum-galactic gas whereas before they were arising more in intergalactic gas. This is supported by Hsiao-Wen's earlier work that most galaxies at z < 1, no matter their luminosity or morphology, have CIV envelopes extending to ~100 kpc. ~~~~~~~~~ Simulations (working paper title and un-finalized author list) ~~~~~~~~~ ~~~~~~~~~ Disclaimer: Just to be honest, I wanted to say that my understanding of the simulations is not deep. I don't have access to all the information yet. But I can share some insights from the simulations, and more will come in the future. ~~~~~~~~~ Why use simulations? Hearkening back to the science drivers and the whole context of the talk, I asked "How many metals are produced?, Where are they? How did they get there?" Simulations are an ideal way to know all this information. But the trick is to determine which simulations reproduce the observed Universe and then understand what about that simulation's physics affect the observations. ~~~~~~~~~ OverWhelmingly Large Simulations Here are the basics of the simulations. They are hydrodynamic cosmological simulations evolved from z = 127 to today with Gadget III (modified version of Gadget II). They have periodic boundary conditions, 2x512^3 particles, and are 100 Mpc/h on a side. For the CIV study, the simulations are converged and the box size is fine. For more on the OWLS project, I refer you to these papers. A slew more are in the works, including two papers by Serena Bertone and one by Marcel Haas (the galaxy images shown previously). Very important for the CIV study is the chemical evolution of the gas. OWLS have a sophisticated treatment of the chemical evolution physics. I refer you to Wiersma et al. for details. I just want to point out that the evolution and effect of 11 elements are tracked in the simulations. The cooling is calculated under the assumption of photoionization equilibrium in the optically thin regime with a spatially uniform UV background (no local sources). Of course, elements are produced and moved around through feedback processes and dynamical interactions. ~~~~~~~~~ The OWLS Philosophy This is taken directly from a presentation by Craig Booth. It nicely summarizes the motivation behind how and why the OWLS were constructed. ~~~~~~~~~ My OWLS "Philosophy" Ok, not so much a philosophy as a modus operandi. We're going to simulate observing CIV absorbers in the OWLS with synthetic spectra. We'll make the same measurements and calculate the same quantities as was done for the observations. Then, remaining firmly in the observational plane, we'll determine which simulation(s) reproduce the observations. We won't be including any noise or blending in the simulated observations. Then, once we found a few observation-reproducing simulations, we'll see how the various physics led to the reasonable simulated observations. ~~~~~~~~~ Synthetic Spectra The synthetic spectra were produced with SpecWizard, which is publically available and is going to be a black box for the purposes of this talk. Please accept that we generate spectra from the simulations that look like (or could look like) the observations. Shown here is the same wavelength swath of an observed spectrum and a simulated spectrum (which has no noise and only Lya, CIV, and SiIV). How we define simulated doublets, though, is not from the flux spectrum but from the apparent optical depth "spectrum." A single CIV profile (or system) is defined to be where the CIV 1548 optical depth consecutively exceeds the continuum optical depth tau_c with at least one pixel greater than or equal to a peak optical depth tau_p. We set tau_c = 10^-3 and tau_p = 10^-2, which is a CIV column density of about 10^11 cm^-2. ~~~~~~~~~ Simulated Observations Data is always good to show. Here are velocity plots of a simulated doublet and an observed doublet. They look about the same. The simulated spectra have higher spectral resolution because to match the highest resolution of the observations. Though we don't include noise when we compile the simulated line list, we show the noise here. They both have S/N = 10 pix^-1. So we can proceed to measure the observables for the simulated doublets. ~~~~~~~~~ Simulated Observations We construct frequency distributions just like for the observations (where I didn't show the f(EW), nor will I discuss it further, just know it's also a constraint on the simulations). Though we don't have any saturated simulated doublets and we can detect absorbers to much lower column densities, we treat the simulated doublets like the observations and fit over the same range, and with the special saturated bin in the maximum likelihood analysis. ~~~~~~~~~ Simulations Description I want you to think of the simulations when I discuss them through the rest of the talk in terms of what they actually are testing. The DEFAULT simulation is the reference simulation, not the one without feedback. It's a carefully tuned simulation that reproduces a lot of other observations pretty well (e.g., star formation history). THERMAL_FB is an outlier from the rest because it doesn't change a few things relative to DEFAULT. A lot has changed because THERMAL_FB represents the "next generation" of OWLS simulations with many improvements. I don't discuss them but just to let you know, we tested the effect resolution and box size and the effect of changing the ionization balance (which approximates the UVB intensity) and abundances (which relates to the yields assumed in the simulations). ~~~~~~~~~ Comparing to Observations: N(C+3) Frequency Distribution Shortly I will page through slides with four plots each, showing information for a sub-set of the simulations, all relative to DEFAULT and the observations. So I want to discuss what you'll be looking at. Here's the main comparison between observations and simulations. In this case, DEFAULT under-produces the observed f(N). However, since the yields are uncertain by that amount, it's conceivable that DEFAULT could be brought into agreement with the observations by changing them. ~~~~~~~~~ Interpreting "Observations": Om(C) and Om(C+3) over Time This is the second plot I'll be showing you. Here I'm showing the total carbon mass density in gas (relative to the critical density) and the total Om(C+3) (from summing up the absorption in all pixels in a redshift slice). The points are the observed 13 <= log N(C+3) <= 15 Om(C+3) and the simulated observed values in two redshift bins. Immediately you can judge how the observed Om(C+3) is not the full story of carbon, or even CIV. [NOTE: These Om(C), Om(C+3) plots have been updated since the defense date.] In DEFAULT, Om(C+3) is 2% to 20% of the total C mass density. But what we "observe" is only 48% of that. And it's the high column density absorbers (which are rare) that make up the other 50%. (Though I'll note that the proportions change for other simulations.) ~~~~~~~~~ Interpreting "Observations": Physical Condition of Gas We've turned to simulations to get a handle on the physical condition of the CIV absorbing gas, which we don't have complete knowledge of through observations (especially the observations we were able to make). One physical parameter from the simulations is the _baryon_ overdensity. Shown here is the contribution to the DEFAULT f(N) at all column densities from gas with different overdensities. The black and white dashed curve is the median overdensity per column density bin. ~~~~~~~~~ Physical Condition of Gas: (Baryon) Overdensity We adopt the following working definitions of what the (baryon) overdensity means. We have the canonical IGM, the diffuse IGM, the CGM, and galactic gas. The IGM-CGM boundary was calculated and just happens to be a nice power of 10. We calculated the overdensity at the virial radius of an NFW halo. And we're assuming that baryons trace dark matter, which we acknowledge is not ideal to have to assume (we should get the actual overdensity) and that the assumption becomes questionable right where we change over from IGM to CGM. This is the third plot I'm going to be showing. ~~~~~~~~~ Interpreting "Observations": Physical Condition of Gas Now we have the temperature of the CIV absorbing gas. Again, here is the fraction of different temperature bins contributing to f(N) for DEFAULT. ~~~~~~~~~ Physical Condition of Gas: Temperature Again, here are our working definitions of what these temperature cuts mean. We have the photoionized gas, the warm gas, the warm-hot gas (where collisional ionization becomes a significant if not dominate factor), and the hot gas (where it's likely all collisionally ionized). Just a reminder, the ionization balance is calculated under the assumption of an optically thin gas in the presence of a spatially uniform UV background (no local sources). The assumption of equilibrium is called into question at log T >~ 4.5 due to shocks. As we can see in DEFAULT, a lot of CIV absorbing gas is at log T > 4.5 so that the cooling rates are not calculated as well as they could be if the simulations allowed for non-equilibrium. This is the fourth plot I'll be showing in following slides. ~~~~~~~~~ Comparing and Interpreting OWLS: DEFAULT Here's the layout of the next several slides. We have the simulated f(N) compared to the observed f(N). From this we gauge if the simulations are producing enough CIV absorption in this column density range and with the right mix of weak and strong absorbers. Down here is the total C and C+3 mass density over time in the simulations. [NOTE: Om(C) and Om(C+3) plots updated since defense date.] Shown is the observed and "observed" Om(C+3), which is limited to systems with 13 <= log N(C+3) <= 15. This tells us whether the simulation has too much (or too little) CIV absorption (shown by f(N)) due to less carbon production and/or due to less carbon in the CIV transition. Up here on the right is the median overdensity (by number) per column density bin for the doublets contributing to f(N). Below 100 is the IGM, above 100 is CGM (and galactic gas). Similarly, in the lower right, is the median temperature per column density bin. Below 10^5 K is photoionized and warm gas, above it is more collisionally-ionized warm-hot and hot gas. Now we're going to page through sub-sets of the simulations. DEFAULT has median overdensity = 176 and median T = 5.6x10^4 K. ~~~~~~~~~ Comparing and Interpreting OWLS: The Basics I call this the basics because it turns out to be the simulations we can rule out (more proof later). Without feedback, there is not enough CIV absorption. Not due to a lack of C (which is actually higher than in DEFAULT) but because of the temperature and/or density (and/or distribution) of the gas. Without metal-line cooling, we also get too little CIV absorption, though less carbon (due to the lower star formation rate in ZCOOL0). The CIV absorbing gas tends to have higher overdensity. The 13 <= log N(C+3) <= 15 absorbers represent 80% of the total Om(C+3). DEFAULT has median overdensity = 176 and median T = 5.6x10^4 K. NOFB has median overdensity = 785 and median T = 1.6x10^4 K. ZXOOL0 has median overdensity = 489 and median T = 7.3x10^4 K. ~~~~~~~~~ Comparing and Interpreting OWLS: Cosmology Since MILL and WML4 have more mass entrained in the winds, they're just directly comparable to each other, though DEFAULT is still here as a reference. MILL has WMAP1 cosmology, which has a higher sigma_8. This means that the simulated Universe evolves faster compared to WML4, which has WMAP3 cosmology. That means more stars have formed in the past, boosting the total carbon mass. The increased star formation due to the higher sigma_8 counteracts the other (small) changes to the physical condition of the CIV absorbing gas due to the higher mass entrained in winds. Because WML4 is less in agreement with observations. DEFAULT has median overdensity = 176 and median T = 5.6x10^4 K. MILL has median overdensity = 160 and median T = 5.2x10^4 K. WML4 has median overdensity = 191 and median T = 4.3x10^4 K. ~~~~~~~~~ Comparing and Interpreting OWLS: Varying Sources Now we're looking at changing what contributes to feedback. Turning off feedback from AGB stars and Type 1a SN decreases the carbon production by 50%. This is mainly due to the lack of AGB feedback; Type Ia SN hardly contribute. Type II SN contribute 70% of the carbon at late times of a single stellar population while AGB stars are about 35%. The physical condition of the gas is not affected much. Remaining C change likely due to change in star formation history (see Oppenheimer and Dave 2008). Including AGN feedback decreases the CIV abundance in large part because, star formation is more efficiently suppressed (in halos where AGN grow). Again, the physical condition of the CIV absorbing gas is not much changed. DEFAULT has median overdensity = 176 and median T = 5.6x10^4 K. AGB0_SNIa0 has median overdensity = 168 and median T = 5.3x10^4 K. AGN has median overdensity = 198 and median T = 6.7x10^4 K. ~~~~~~~~~ Comparing and Interpreting OWLS: Wind Model In WML4, the wind entrains twice as much mass as in DEFAULT. However, the material may not get very far in massive halos because the wind model has a constant initial velocity. Still, star formation is suppressed in halos of all masses, so that less carbon is produced. The physical conditions of the gas do not change. WMOM uses a momentum-driven wind prescription that scales the initial wind velocity with the halo mass. This is sort of a gentler wind model for < 10^11 solar mass halos, and star formation is greater than in DEFAULT. However, above about 10^11 solar masses, star formation is more efficiently suppressed (and indeed, the sample galaxy shown earlier was completely disrupted in the WMOM run). However, the CIV absorbing gas is typically hotter (70%) in WMOM than in DEFAULT, which is perhaps why the CIV absorbers are less numerous. DEFAULT has median overdensity = 176 and median T = 5.6x10^4 K. WML4 has median overdensity = 191 and median T = 4.3x10^4 K. WMOM has median overdensity = 241 and median T = 9.5x10^4 K. ~~~~~~~~~ Comparing and Interpreting OWLS: "Next Generation" OWLS I'll emphasize again that it's very difficult to know exactly why THERMAL_FB differs as it does because several things were changed. But, with that, THERMAL_FB reproduces the observations the best. Finally, there is enough CIV absorption in the simulation. I don't have the carbon mass for this run, but since sigma_8 is slightly higher, the carbon mass might be slightly higher. The total CIV mass density does not differ much from DEFAULT, though the observed 13 <= log N(C+3) <= 15 Om(C+3) agrees well with the observations (more so at high redshift or taken over the full z = 0 to 1). The median overdensity (over all column densities) is about 50% of DEFAULT's. This is due to the increased number of low column density absorbers. We can see the turnover because 13 <= log N(C+3) <= 15 traces about 60% of the mass density, where as log N(C+3) > 15 is only 37%. Very different from DEFAULT, which was 48% and 50%, respectively. But THERMAL_FB also has more high N(C+3) doublets (slope fit). DEFAULT has median overdensity = 176 and median T = 5.6x10^4 K. THERMAL_FB has median overdensity = 96 and median T = 5.0x10^4 K. ~~~~~~~~~ Comparing to Observations: N(C+3) Frequency Distribution Fits I like looking at the best-fit power-law parameters and their error ellipses to distinguish between the simulations. THERMAL_FB stands out as clearly the best. AGB0_SNIa0, NOFB, and ZCOOL0 are ruled out with high confidence (other test than that shown here, like two-sided KS tests). DEFAULT and MILL do well and could even be brought into agreement by changing the yields (i.e., the carbon abundance). "Just need feedback": I say this because, if it's really true about the yields being flexible, then you get several different sub-resolution feedback models producing results in agreement with observations. So whatever might be happening elsewhere, the CIV bearing gas just needs feedback that keeps/puts material in regions with overdensity > 100 and T ~ 10^4.5 K. Caveat: looking at the error ellipses and using them to judge assumes that the power law is a good formalism, which is not the case for all simulations. But what I said about these three stands. ~~~~~~~~~ Comparing to Observations: dN/dX over Om(C+3) There are other observables we can use to distinguish between simulations. Still THERMAL_FB looks the best, as shown earlier. ~~~~~~~~~ Comparing to Observations: dN/dX and Om(C+3) Evolution There is also the evolution of the observed quantities to judge between simulations. But we need better data because there is barely enough absorbers in each redshift bin to constrain the values. And the observed values are in agreement with each other. ~~~~~~~~~ Physical Condition of Gas To summarize briefly, 13 <= log N(C+3) <= 15 Om(C+3) dominated by circum-galactic gas (based on the baryon overdensity in the simulations). It seems like pretty much all log N(C+3) > 14 doublets are in galaxy halos. The simulations that are not completely ruled out (and sort of ignoring WMOM) have CIV absorption arising in photoionized gas. ~~~~~~~~~ Column Density Maps Let's zoom in on a 500 kpc thick slice of the DEFAULT simulation to see what the column density "structure" is. These are beautiful images made by Serena Bertone. The observed column density range (blue to pink) traces galaxy halos. ~~~~~~~~~ Column Density Maps The halos are larger than the 100 kpc Hsiao-Wen measured in her 2001 paper, but I don't have the complete information of what exactly this object is. (Serena says they're galaxies.) ~~~~~~~~~ Simulations Summary We have a winner in THERMAL_FB. So I'm eagerly awaiting the future OWLS to compare and contrast with. But from the standard OWLS, the DEFAULT run does quite well. It may even be as good as THERMAL_FB if re-run with higher yields. The most reasonable OWLS indicate that the observed CIV bearing gas is circum-galactic (meaning its baryon overdensity is > 100) and photoionized. ~~~~~~~~~ Future Work ~~~~~~~~~ ~~~~~~~~~ SiIV: Wash, Rinse, Repeat The data is primed and ready for me to search for SiIV, which is another fairly well studied metal in the IGM (see Anthony's 2004 paper). In its own right, SiIV absorption is interesting, just like CIV. In combination with CIV, we can look at e.g. the shape of the UVB. It will also add another dimension by which to judge the simulations. ~~~~~~~~~ Other Future Work Here's a brain dump of the top things to do next. Let's get more serious about what is halo gas and what is not by masking out non-halo gas from the simulated observations. COS is going to be great for increasing the statistics so we can look at the evolution of the metals (CIV and SiIV). I know I haven't treated the simulations like the observations. I stated that earlier. It's necessary to now go back and do that: add noise, add patchy redshift coverage, varying spectral resolution, and even allow blending. I didn't discuss SpecWizard, but I can use it more effectively. I also have recommendations for the OWLS team (who definitely have a lot on their plates right now so I'm not going to harass them... as a matter of fact, I have enough to keep me busy for a long time.) ~~~~~~~~~ Wrapping Up ~~~~~~~~~ ~~~~~~~~~ C+3 Mass Density: Over Time This is a good summary image. Observed 13 <= log N(C+3) <= 15 Om(C+3) has increased almost threefold. Some simulations do well to reproduce the observations, but the redshift evolution is an important constraint, which needs to be nailed down observationally. ~~~~~~~~~ Summary Repetition is my friend. Om(C+3) increases statistically significantly compared to 1.5 < z < 5 error-weighted average. Something conspires to keep dN/dX changing little from high to low redshift. CIV absorbers are CGM more than IGM. ~~~~~~~~~ Thank you... The incomplete list of who I'm grateful for... ~~~~~~~~~ Summary I'll take questions. ~~~~~~~~~ ~~~~~~~~~ Preview: C+3 Mass Density Though I've worked on other metals, I'm going to be focusing on carbon today, specifically CIV 1548, 1550. As mentioned previously, a lot of work has been done on the z > 1 IGM. This includes CIV. Specifically, measuring the mass density, which is plotted here relative to the critical density: Omega(C+3). These are only a sample of the Omega(C+3) measurements from z > 1. For example, I've excluded Anthony and Joop Schaye's work using the pixel optical depth technique. I just stuck to looking at traditional absorption line studies, though the POD studies show the same result. So the interesting aspect of this plot is that Omega(C+3) has not changed (much or at all) from z = 5 to z = 1.5. ~~~~~~~~~ Preview: C+3 Mass Density If I clean up the plot and take the error-weighted average of these redshift bins, it's more apparent that there has been little to no evolution from z = 5 to z = 1.5. ~~~~~~~~~ Preview: C+3 Mass Density However, and I thank Hsiao-Wen for suggesting that I make the plot like this, all that business of z = 5 to 1.5 is 3 Gyr. There is almost 8 Gyr for something to happen. Before our paper, there had been two studies on Omega(C+3) at z < 1. Our study is bigger and better. "We" refers to Chris Thom at U Chicago (now off to Space Telescope), Xavier, Hsiao-Wen, and myself. So let me describe it. ~~~~~~~~~ Evolution of the Cosmic Web Here's a quick overview of the evolution of the Universe with highlights that are important to metals in the IGM. In the beginning, there were (nearly) no metals. The small density fluctuations grew, so that stars started forming, making metals, and dying, dispersing the metals. And some metals are produced as some stars die. The overdense regions keep growing into the filamentary structure we "see" today. Stars keep forming and dying, all the while creating and releasing metals. Star formation starts to decrease at z ~ 2, which will effect the creation and release rates. More is known about the first ~5 Gyr of the cosmic web properties than about the last 8 Gyr. ~~~~~~~~~ Example CIV: GHRS Detections Here are some examples of how we recovered known absorption systems based solely on a CIV-targeted search. This also shows me that GHRS was worth the effort. ~~~~~~~~~ C+3 Mass Density: Over Time Here's the big result (and several people have seen the "answer" already).