The Star-formation History of the Universe

Alan Heavens. American Scientist. Volume 93, Issue 1. Jan/Feb 2005.

More than nine thousand billion billion (9 × 1021) stars have been formed in the observable universe since it began 13.7 billion years ago. Despite the apparent wealth of stars in the sky, current cosmological models suggest that the universe was quite dark for much of its first billion years. During these dark ages the universe contained clouds of gas and dark matter, but little else-the first stars did not form until several hundred million years had passed. Once the cosmic star-making machinery got going it seems to have churned out stars at a prodigious rate. But was this rate constant throughout the history of the universe? Or was there anything akin to a baby boom?

The answer provides more than just another curious bit of information about the cosmos. Just as the postWorld War II baby boom was an indicator of profound social and economic changes in the West, differing rates of star formation provide clues about the physical circumstances in which star birth takes place. These “physical circumstances” are, of course, the galaxies, and the rate at which stars were made is intimately related to how the galaxies were formed.

It so happens that we can come up with some reasonable answers to these questions with the aid of modern telescopes and computers. My colleagues and I have recently made considerable strides in determining the history of star formation in the universe by analyzing the starlight from 96,545 galaxies. The results are surprising: On the surface they seem to contradict some basic principles of how the universe assembles a hefty structure like a galaxy. The paradox can be resolved if we consider that the galaxies weren’t all made the same way.

The Universe Makes Structures

Ever since 1917, when Vesto Slipher noticed that the spectral lines of a galaxy’s light emissions are routinely shifted toward longer (“redder”) wavelengths, we have had firm evidence that the universe is expanding. The nature of this expansion depends on how much matter the universe contains, and cosmologists have been trying to determine its density for decades now. Recent observations of exploding stars, the largescale structure of the universe and the fireball radiation left over from the Big Bang-called the “cosmic microwave background”-have shown us that the universe does not have enough matter to stop the expansion. Furthermore, the matter it does contain appears to be made of some rather strange stuff. Only 4 percent of the density is made up of ordinary matter-neutrons, protons, electrons and so on-the kind that makes up people, planets and pixels. Another 23 percent is dark matter, an extraordinary and unknown material that is not seen on Earth. And 73 percent is in dark energy, an even more bizarre substance that has a “repulsive gravity,” which is causing the universe to accelerate its expansion.

Despite the admittedly mysterious circumstances, we know quite well in broad outline how the universe formed the structures we see. Observations of the cosmic microwave background, which dates to when the universe was only 300,000 years old, show that the universe was not quite uniform in its early phases. These observations have been made with a number of ground-, balloon- and satellite-based experiments, most famously with the Cosmic Background Explorer in the early 1990s, and now at higher resolution with the Wilkinson Microwave Anisotropy Probe. These experiments revealed small irregularities in the density of the early universe. The irregularities also happened to be unstable: Denser-than-average regions had slightly stronger gravity, and so pulled matter in to form “clumps.” In this way dense objects formed over time through a combination of gentle accretion and the merger of smaller units.

Ironically, to form something hot like a star, gas has to cool first. Large clouds of gas can support themselves by pressure if they are hot, but if the gas cools the pressure is removed and the cloud collapses. On the smallest scales, the final stages of collapse will heat the cloud, and if the temperature reaches some millions of degrees, nuclear reactions may start and the object becomes a star. On large scales, we believe the process of collapse operates bottomup: Relatively “small” objects (around a million times the mass of the sun) collapse first. As the dense gas cools, it fragments into small clouds that ultimately collapse into stars.

These systems of many stars are drawn together by gravity into larger systems, and these larger systems themselves may coalesce. And so the process goes, building larger systems in a hierarchical fashion, through small galaxies, large galaxies, galaxy groups and finally galaxy clusters and superclusters. Today the largest structures that have fully collapsed are clusters of galaxies containing dark matter, gas and star-filled galaxies that total some hundreds of thousands of billions of solar masses.

This hierarchical picture of galaxy formation allows us to understand many things-how spiral galaxies spin, why we observe a wide range in the masses of galaxies, and even why there is a limit to the largest objects we see. However, many details are sketchy. The formation of a galaxy is a complex process, and the way that galaxies turn their gas into stars is not fully understood. For example, stars can influence the gas through supernova explosions and winds in a “feedback” process. Feedback may enhance or suppress star formation, but reliable estimates of the extent of these processes still elude us.

It is within this setting that some astronomers and cosmologists, my colleagues and I included, are attempting to determine when the universe made most of its stars, and how this rate may have changed with time.

Fossils vs. Newborns

The problem of working out how many stars were born over time is analogous to trying to determine the historical birth rate in the human population. One could either look into the historical records and identify all the births and their dates, or take a snapshot of people and estimate their ages. With the latter approach, one has to be careful to include all the places that people could be, such as a graveyard. In the former case, it is important to have a complete historical record. Of course, if nothing has gone wrong, these two methods should agree.

Similarly, the history of the starformation rate of the universe can be determined in two different ways. The traditional way is to use the finite speed of light to look back in time. By looking at the light from very distant objects, we see them as they were many billions of years ago. One can then search for signs of newborn stars at that time. The alternative is to use a “fossil” approach. Here one looks at a very large sample of nearby galaxies and measures the ages of the stars in each galaxy. This is the approach that I used in close collaboration with my colleagues, Raul Jimenez at the University of Pennsylvania, and Ben Panter and James Dunlop of the University of Edinburgh.

Until recently, the study of the starformation history of the universe has been dominated by looking for signs of newborn stars over time-reading the “historical records” of star birth. Typically, this has been done by observing distant, or “high-redshift,” galaxies. One indicator of recent star formation in a galaxy is the amount of ultraviolet light it emits. Ultraviolet light typically comes from very hot stars, such as those seen in the Pleiades star cluster. Because very hot stars are massive, and such stars burn up their nuclear fuel very fast arid then die, we can only see them when they are very young. The presence of ultraviolet light is thus a good gauge of newborn stars in a galaxy. On the other hand, as the stellar population gets older, the bright blue stars become redder, and the starlight takes on a reddish or yellowish hue. There are other signatures of recent star-formation activity, such as the emission of spectral lines of hydrogen, which comes from ionized hydrogen gas around massive stars. These stars produce enough ultraviolet light to ionize large volumes of surrounding gas, which then glows. Studying high-redshift galaxies is a time-consuming task, however, and is limited to observing only the most massive systems because smaller ones are simply too faint to see at great distances.

The fossil approach is quite different. This method uses the spectrum of light from a galaxy over a broad range of wavelengths. Most of this light comes from the individual stars in the galaxy, but some is emitted by hot glowing gas. By removing the bulk of the gas emission lines, we leave mostly starlight. Each star contributes differently to the spectrum, depending on its age and the chemical composition of the gas from which it formed. Since the elements are made inside the stars, the heavier elements-which astronomers call metals-accumulate in the gas and the younger generations of stars as a galaxy ages. A careful analysis of the spectrum can tell us the age of the stars and how many stars were formed at different periods in the galaxy’s history. We can also deduce how the chemical composition of the gas changed through time.

Consider the spectrum of a galaxy that contains some young stars, including some massive stars that emit a lot of blue light. This is evident as relatively high emissions at “blue” wavelengths, in the vicinity of 400 nanometers. In contrast, the spectrum of another galaxy shows very little blue light, indicating that hot, massive stars are no longer burning bright. Such a galaxy lias had very little recent star formation-the population is old, and the light is dominated by cooler red stars. By analyzing a large number of such spectra, we can build up a picture of how much star formation took place at a particular time in the universe’s history.


The Sloan Digital Sky Survey (SDSS) is a tremendously valuable resource for our work. The SDSS will eventually map nearly a quarter of the sky, measuring the spectrum of many hundreds of thousands of galaxies up to three-billion light-years away. The American-led survey uses a dedicated 2.5-meter telescope at Apache Point in New Mexico. The telescope is connected to 30 charge-coupled devices (CCDs), which produce as much as 200 gigabytes of imaging data per night. The spectra of 640 galaxies are obtained simultaneously in a single “snapshot” by sending their light down optical fibers to two spectrographs.

The process of decoding a spectrum is not easy. To ease the calculations, we divide the history of a galaxy into 11 time periods. For each period, we determine the fraction of stars in the galaxy that were born at that time. We also assess the chemical composition of the galaxy for each of those periods. The issue is complicated somewhat because many galaxies contain dust, which tends to absorb blue light, leaving a galaxy looking redder than it really is. So, with 11 periods of star formation, 11 measures of chemical composition and one measure of dust, we have 23 numbers to find for each galaxy.

One way to determine these numbers is to match the observed spectrum with a model spectrum that provides a close fit. This task is possible in principle, but far too slow in practice, especially when one takes into account that we have nearly 100,000 galaxies to analyze. Instead, we use a massive data compression algorithm-called MOPED (Multiple Optimized Parameter Estimation and Data compression)-which reduces each galaxy to 23 numbers instead of several thousand points on the spectrum. This makes the analysis go nearly a hundred times faster-without compromising accuracy. MOPED is a patented technology that we developed for this task, but it is applicable to a wide range of inverse problems like this. Nevertheless, it still requires several weeks using 20 computers to process the whole survey.

MOPED provides the complete star-formation history of each galaxy. If we add up all the galaxies in the survey appropriately, we can measure the average star-formation rate per unit volume in the universe. Our results agree with those of the more established methods at early times and at recent times in the universe’s history, but they disagree in the middle. The traditional methods suggest that the star-formation rate in the universe peaked about 8 billion years ago, but the new results show that it peaked about 5 billion years ago, around the time that the sun and the solar system were born. Both methods show that star formation has been in a fairly steep decline, so that nowadays the rate is only about 10 to 15 percent of what it was at the peak. The new peak we found using the fossil approach should be confirmed by the other methods, but this will not be easy to do because it is difficult to see all but the brightest galaxies at the great distances involved.

So how do we explain the three-billion-year discrepancy? A big clue to reconciling the differences comes from looking at the star-formation rates of galaxies of different masses. There is a very clear trend: Massive galaxies have an earlier peak in star-formation activity. That is to say, massive galaxies, such as our Milky Way, formed most of their stars early and by now have more or less stopped making stars. Indeed, current estimates suggest that our galaxy now makes only a handful of new stars each year. Low-mass galaxies, on the other hand, form most of their stars late, and continue to make stars at a good rate even today.

The key here is that the traditional methods, which were based on observing distant galaxies, can only detect the most massive ones. As we have seen with the galaxies in the Sloan survey, the most massive galaxies are unrepresentative in that they formed most of their stars unusually early. So it is no surprise that the previous determinations found an early peak in star-formation activity. In the analogy with the human birth rate, it is as if the small parish records were lost, and all we had to go on were the records in the cathedrals. All is well and good if the cathedrals are representative, but if demographic changes meant that the cathedrals recorded a progressively smaller fraction of the births, their records would give us an inaccurate picture.


How does all of this fit into our standard picture of galaxy formation? As first sight, it appears to fly directly in the face of the hierarchical model, which states very firmly that small objects should form first and large ones later. We appear to be seeing the opposite: The big galaxies are forming their stars before the small ones. There is, however, one other important fact about our study of the SDSS galaxies. We look at stars that are in the galaxy today and work out when they were born. We have no direct knowledge of where they were born. In particular, there is no guarantee that the 8-billion-year-old stars in a galaxy were actually in the galaxy when they were born. They could have been made in several different smaller galaxies, which merged together at some point in their history. In fact, that probably is what happened: Most of the stars in the massive galaxies were almost certainly made in smaller units, which were later assembled to form the big galaxy.

The low-mass galaxies probably do this the other way around. The mass is largely assembled first, and then it pulls in gas from its surroundings. The infalling gas fuels further star formation, or even triggers star formation in the gas that is already there. One can go even further and say that, ironically, in the bottom-up hierarchical model of galaxy formation, we expect the objects that end up being the most massive to form their stars early!

One interesting consequence of having two methods to study this question is that we can put the Copernican principle to the test. The Copernican principle says that we don’t inhabit a special location in the universe. Since MOPED tells us what the star-formation rate was here in the past, and the other methods tell us what the star-formation rate was a great distance away, we can compare the two to see if our location is unusual. Since we believe we understand the different histories that we do see between the two methods, we can conclude that the Copernican principle holds true in this instance.

So we know that our universe experienced a baby boom in star formation roughly 5 billion years ago. Is there any hope for another boom in the future? Stellar obstetricians will be saddened to learn that the future looks pretty dim. We are well past the heyday of star formation, and the rate is certainly in decline. As progressively larger structures form in the universe, the infalling gas is getting heated to ever higher temperatures, which makes it very difficult for the gas to cool and collapse to form new stars. The gas in the universe is also generally getting thinner with time-the supply that is suitable for making stars is “drying up.” We are faced with a future of skies that are becoming ever darker. But this is not something that we, or our grandchildren, need to fear-our Sun has another 5 billion years to go.