Back in June last year, a nice little paper showed that it was possible to use stacking to measure the atomic gas content of very distant galaxies indeed. Normally, directly measuring the atomic hydrogen content of galaxies becomes excessively difficult beyond the relatively local Universe. Beyond a couple of hundred megaparasecs it becomes really quite challenging; to reach a gigaparsec is just about possible, but at the extreme limits of technical capabilities. The 21cm emission from hydrogen is just too bloody faint.
So the paper used a standard technique of "stacking" the detections : observing lots and lots of galaxies and averaging all their signals together. In this way, you get the sort-of-but-not really equivalent to spending an insane amount of observing time on an individual galaxy (see link for subtleties). The previous paper didn't do much but establish the main result, which is fair enough, but in this new paper they've gone quite a bit further.
Using new, better data for over 11,000 galaxies, they can now recover a signal from a redshift z=1.4, equivalent to a lookback time of 9 billion years. This is well in the range where we'd expect to see evolution in the typical gas content. Last time they established that, as expected, earlier galaxies have a higher gas fraction than their nearby contemporaries, but now they do two additional things. First, they can split their sample to check the gas content at z=0.74-1.25 and for z=1.25-1.45. Adding a separate sample of very nearby objects means they can plot the evolution of the gas content with no less than three, count 'em, three data points - enough, some would say, to see a trend. Second, they also estimate the molecular gas content, so they can compare how the atomic and molecular fractions evolve.
I'm going to gloss over some pretty hefty caveats. First, not only is stacking itself a subtle and oft-misunderstood procedure, but it's also subject to a host of biases and selection effects. However, their detections are convincing, so I note that only for caution, not as criticism. Potentially more problematically, their estimate of the molecular gas comes not from direct measurement but relationships between star formation rate and molecular gas content established elsewhere. As they note, this too may be biased.
However, let's give them the benefit of the doubt and take their results at face value. It seems that the atomic fraction has been monotonically decreasing over time, dropping rapidly from z=1.4 to 1.0, but then more slowly to the present-day z=0.0 (with the reminder that this is based on three data points). The molecular gas fraction is more complicated : it's lowest at z=0.0, peaks at z=1.0, then decreases a bit at z=1.4.
What's going on here ? Well, the prevailing view is that atomic gas the reservoir of fuel for star formation, but it's molecular gas which is actually where star formation happens. If atomic gas is what's in the tank, then molecular gas is what's in the engine. So we might naively expect the gas to be predominantly molecular during the peak of star formation activity. The problem is that's not what we see here : star formation peaks when the gas is dominated by atomic gas instead.
That's a bit strange, but I wouldn't be overly-concerned that everything is broken just yet. Conversion of molecular gas to stars involves a great deal of highly complex feedback, where the hot young stars and supernovae inject energy back into the gas and helping to slow further conversion into molecular gas and thus stars. It looks like the early rapid drop of atomic gas is balanced by the corresponding, combined increase in molecular gas and stars, so that fits the basic narrative. The hard bit is explaining why star formation activity was higher at earlier times, when there apparently wasn't all that much molecular gas around at all.
There's been some hints that the basic scenario is too simple, that actually the atomic gas itself can be directly involved in star formation. That could well be the case. No galaxy, so far as I'm aware, in the nearby Universe, shows atomic gas fractions as high as the ones at high redshift described here - so things might just work differently. Alternatively, it could be that the molecular content has been underestimated, or that the statistical effects of all this stacking are washing out a plethora of important details. Unfortunately, we're probably not going to be able to remedy this without direct detections of atomic hydrogen as z=1, which isn't going to happen anytime soon... though it's likely to be years (rather than decades) away. So, we'll see.
