The discovery by separate teams of scientists a decade ago that the expansion of the universe is accelerating was one of the most important events in cosmology in recent years. The chronology of the competition between the two teams — and the role that the rivalry played in the ensuing events — illustrates key ambiguities in features of the scientific process, including the nature of discovery, the way in which announcements are made and the difficulty in assigning credit.

Making the discovery

The story of the discovery of the accelerating universe begins in 1987 when physicists at the Lawrence Berkeley National Laboratory and the University of California at Berkeley initiated the Supernova Cosmology Project (SCP) to hunt for certain distant exploding stars, known as Type Ia supernovae. They hoped to use these stars to calculate, among other things, the rate at which the expansion of the universe was slowing down. Deceleration was expected because in the absence of what Einstein called a cosmological constant — an antigravity force, Λ, that pushes matter outwards — many people thought that ΩM, which is the amount of observable matter in the universe today as a fraction of the critical density, was sufficient to slow the universe's expansion forever, if not to bring it to an eventual halt.

Saul Perlmutter, who had worked on robotic methods of searching nearby supernovae as part of his PhD dissertation, eventually became head of the team. His colleagues included Gerson Goldhaber, another physicist well versed in particle physics imaging techniques. Spotting the ephemeral supernovae — which appear unexpectedly and then wink out a few weeks later — is extremely difficult. Indeed, at about the same time researchers in Denmark led by Hans Nørgaard-Nielsen spent two years intensively searching for distant Type Ia supernovae but managed to find only one example, 1988u. Moreover, that supernova was already several weeks past the crucial peak brightness that is used to calibrate supernovae — and so could not be used for the desired kind of measurement. Independently, the SCP team, which then numbered less than half a dozen but the personnel of which would vary over the years, had developed a number of new experimental approaches to address the challenge and get its project going. These included a novel widefield camera that could observe thousands of very distant galaxies in a single night.

Unlike the Danish researchers, the SCP team searched for supernovae using a "two-visit" method, which involved imaging the same area of the sky twice between two new Moons. The second observation was designed to uncover a batch of newly brightening supernova candidates not yet present in the first observation; the team could then schedule follow-up time on other telescopes to measure these supernovae in detail. The SCP scientists introduced other innovations as well. For example, while it was common for astronomers to use a logarithmic magnitude scale for brightness, the SCP team used the actual units of flux and energy in its statistics. The team also devised an improved method of "K-corrections" for supernovae, in which different kinds of filters were used to catch red- and blue-shifted light. This method had occasionally been used in the past to study galaxies, but has now been adopted by all supernova teams.

Initially, the members of the SCP team, who were relative newcomers, had trouble, in the fiercely competitive astronomical world, securing follow-up time on oversubscribed premier telescopes such as the Cerro Tololo Observatory in Chile. The team spent several years fine-tuning its approach, struggling with bouts of bad weather and resistance from programme committees to devoting telescope time on an unproven technique. For follow-up spectroscopic measurements, the team sought and obtained help from astronomers with already approved programmes. In 1992 the SCP researchers finally found a supernova — 1992bi — the most distant one then discovered, and submitted a paper. It would become the first publication on a distant supernova that had been spotted early enough to measure its peak brightness.

But not immediately. Its reviewer, Harvard astrophysicist Robert Kirshner, thought he saw conceptual flaws in the paper. Among other things, he disagreed with the SCP team's approach to handling dust. (Dust makes supernovae appear dimmer and therefore more distant than they really are, and can distort conclusions about the cosmology.) The SCP team's paper (Astrophys. J. 440 L41) was not published until 1995, when the team finally convinced a second well-known senior reviewer, Allan Sandage, that its techniques, including the novel and now-standard way of handling K-corrections, were trustworthy, and he allowed the team not to worry about dust for now.

By 1995 the SCP group had produced regular batches of supernovae candidates, found and measured seven, and had demonstrated the ease and effectiveness of the approach. Brian Schmidt (a former student of Kirshner) took notice, appreciated the difficulties that the SCP team was having with dust, thought that there were better ways of handling it than those to which the SCP team was committed, and recruited several other astronomers to form another collaboration, called the High-Z Supernova Search Team. "Continuing unease with the dust issue," Schmidt told me, "more than anything, led to the formation of the High-Z team." Schmidt has also written that the team decided to use its "expertise in understanding and measuring supernovae as [its] competitive advantage", especially with respect to the tricky dust question. Other leading members included Nicholas Suntzeff, then at Cerro Tololo, Kirshner and his former student Adam Riess, who studied interstellar dust for his dissertation. Alex Filippenko, a world expert on the spectral analysis of supernovae, soon left the SCP team to join the High-Z team. The High-Z astronomers promptly secured time on the Cerro Tololo telescope. Still, the High-Z team was way behind, and Kirshner even wondered if it was too late to compete with the SCP researchers.

Its competitors didn't mind. As Goldhaber puts it, "The sky is a big place".

The SCP team submitted a paper describing its techniques and data on the first batch of seven supernovae (1997 Astrophys. J. 483 565). Because the sample was small, the data had a large error bar. It pointed to a value of ΩM in the universe of 0.88. At face value, this was tantalizingly close to a then-prevailing, relatively simple theoretical model, whereby ΩM = 1, which would indicate that the universe is on the knife-edge of recollapse. The researchers also knew that the path to the final answer lay in acquiring more statistics, and noted in their paper that another 18 supernovae were being measured. "Once again, the analysis of our next set of high-redshift supernovae will test and refine these results," the paper concluded.

The SCP team soon included more supernovae in its analysis, including 1997ap, the most distant supernova yet detected. Because 1997ap was so far out and cleanly measured by the Hubble Space Telescope, it powerfully affected the analysis. When the SCP team incorporated it into a paper that was published in January 1998 (Nature 391 51), the data suggested a value for ΩM that was lower than 0.88. As a related commentary in that issue of Nature noted, such a value would imply the existence of a cosmological constant, if the standard inflationary theory of the start of the universe were correct. On 7 October 1997 — a week after the SCP team submitted its Nature paper — the High-Z team independently submitted its first paper, the lead author of which was Peter Garnavich. That paper, which was published in February 1998 in the Astrophysical Journal (493 L53), also gave a low value for ΩM.

Two quite different groups, with contrasting styles and traditions of expertise, were now hunting distant supernovae — and beyond that, the rate of deceleration of the universe. Each group worked feverishly, aware of the other's presence; sometimes they even used telescope time back to back. Yet because the supernova community is small and close-knit, interaction was inevitable. "We shared each other's good ideas," says Schmidt. The High-Z team helped the SCP team by swapping telescope time during a scheduling crisis, and the SCP team helped the High-Z team by measuring spectra during a period of bad weather, and providing pre-publication K-correction calculations. Indeed, several publications from the time include members of both groups as authors. As a benchmark, both groups used a valuable database of low-Z Type Ia supernovae that had been compiled by a team of Chilean and US astronomers known as the Calan/Tololo team — all the members of which (including Suntzeff) had since joined the High-Z team. Meanwhile, several members of the former Danish team joined the SCP team.

The ambiguities had just begun.

Announcing the news

By the autumn of 1997 the SCP team had built up its statistics, having found 40 distant supernovae. Team members worked round the clock on various stages of the analysis. Taking an initial look at the implications for cosmology, in his notebook Goldhaber plotted supernova magnitudes against redshift, creating what is known as a Hubble diagram. He noticed — as others on the team had also begun to suspect — that the data implied a negative value for the mass density, ΩM. This was a nonsensical result that could only be made meaningful by introducing a cosmological constant — an outwardly pushing force. This in turn implied that the universe is accelerating.

Although the data in the SCP team's Nature paper, which at the time was at press, contained hints of a cosmological constant given the scenario of an inflationary universe, now the data bore evidence of a cosmological constant whether or not we live in an inflationary universe. This implication — which ruled out the interpretation of its earlier paper — underscored the concern that the SCP team's paper had expressed about interpreting too much from a small sample size.

Meanwhile, in the High-Z team, Schmidt led the effort to search and discover supernovae, with Riess now leading calibration and analysis of data. Riess's lab notebooks from autumn 1997 show that he, too, was seeing signs of cosmic acceleration in the form of negative mass, and High-Z members also produced a preliminary Hubble diagram with the objects in accelerating territory. In public presentations, though, the team remained agnostic, saying "We don't know!" on the question of whether the universe was speeding up or slowing down.

Then a sequence of events unfolded, the meaning of which is still hotly debated by certain individuals in the two groups.

Goldhaber showed his preliminary Hubble diagram at an SCP group meeting on 24 September 1997. Sceptical but galvanized, his colleagues began cross-checking details. Perlmutter and Goldhaber gave colloquia on their current results. On 8 January 1998 at a meeting of the American Astronomical Society (AAS) in Washington, DC, the SCP team made the first large-scale presentation of its data to the scientific community. The team gave a talk, held a press conference and exhibited a poster.

The poster (put online a few months later at www.supernova.lbl.gov and on 30 December 1998 to the arXiv.org preprint server) discussed the techniques and dataset. It included a Hubble diagram as well as a graph of ΩM versus ΩΛ (the energy density of the cosmological constant) that contained a bull's-eye-like shape that showed clear evidence of a cosmological constant. In the worst-case scenario, the diagram still showed the possibility of no cosmological constant only as a small tail, in the odd case of a universe with highly curved space and very low mass. The poster reads, "note that the confidence regions do not include the 'standard model' inflationary universe with no cosmological constant". It also cautioned that "the dashed line confidence region on the right plot shows our preliminary estimate of this systematic uncertainty...Further analysis should reduce this uncertainty".

The team's press release — entitled "Distant exploding stars foretell fate of the universe" — focused on the discovery that the universe would not recollapse. At the full press conference, a theorist was asked to come and explain to journalists what a cosmological constant was and why it was important. A few reporters got the message. One was James Glanz of Science, who wrote an article entitled "Exploding stars point to a universal repulsive force". Another was Charles Petit, whose front-page story in the next day's San Francisco Chronicle was headlined "Scientists see cosmic growth spurt" and whose story "A few starry and universal truths" in US News & World Report a few days later wrote of the astronomers' reports of the "accelerating expansion" of the universe.

"If it was obvious to me, it was obvious to every cosmologist," Petit told me. "I was very surprised that it did not become a bigger story." However, other reporters did not recognize what Petit had, and only reported the news that the universe would expand infinitely. Presenting its systematic uncertainty may have scared off reporters that a final answer may not have been reached. As Glanz quoted Perlmutter as saying, "[He] cautioned that the group was still correcting for possible dimming of the light by dust and that the conclusions could still change."

Meanwhile, Schmidt and Riess had been talking about their preliminary results. They completed their analysis on 8 January 1998, the day of the AAS meeting — not in time to make an announcement, and not knowing what the SCP team would say. A few days after the meeting, the High-Z researchers began discussing their data as an entire team. Riess, who got married on 10 January 1998 and had been analysing data on his honeymoon, e-mailed the rest of the team two days later that "the data require a non-zero cosmological constant! Approach these results not with your heart or head but with your eyes", adding that "the results are very surprising, shocking even". Aware of the competition, the High-Z team kept its conclusions to itself pending a formal announcement — wary, Riess says, of crying "Wolf!". Riess joked in his e-mails of "the LBL guys running around" and urged his collaborators "to work carefully and efficiently, and maybe the tortoise can catch the hare".

The High-Z team's spirits partly rested on self-confidence in its ability to handle dust corrections — Riess's 1996 dissertation on the subject had won an award — and in its suspicion that the SCP team still did not have an adequate strategy to handle the problem. The team also had a smaller dataset of supernovae to analyse: 10 fully analysed and four "snapshot" supernovae (ones with weaker data), plus two previously published from the SCP team's work, including 1997ap (though in the subsequent publication it was not clearly identified as such).

On 18 February at the Dark Energy Conference in Marina Del Rey, California, Goldhaber and Perlmutter discussed the SCP team's findings. They were followed by Filippenko, who had left the SCP group a few years before. Filippenko announced the High-Z group's conclusion that the universe was expanding at an accelerating rate.

The tortoise (High-Z) did indeed catch the hare (SCP) as far as publications were concerned. Less than a month later, the High-Z team submitted a detailed paper on its 16 supernovae (10 fully analysed, four snapshot and two SCP supernovae) with the unequivocal title "Observational evidence from supernovae for an accelerating universe and a cosmological constant" to the Astronomical Journal. The paper was refereed, posted online in May and published in September 1998 (A G Riess et al. Astron. J. 116 1009). The SCP took longer to complete all the analyses on their 42 supernovae, and submitted a paper on 8 September 1998. Following a physics tradition of giving deadpan titles to papers announcing dramatic discoveries, it was entitled "Measurements of omega and lambda from 42 high-redshift supernovae". It was posted online in December 1998, and published in June 1999 (S Perlmutter et al. Astrophys. J. 517 565).

Awarding credit

The combined evidence for a cosmological constant from the two teams convinced most of the scientific community with a speed that is remarkable for a dramatic and unexpected new discovery. In March 1998, at Fermilab, members of the two groups presented their case; and at the end, audience members were asked to raise their hands if they believed a cosmological constant existed. By a vast majority, they did. In December 1998 Science magazine selected the work as its breakthrough of the year. In 2001 a (methodologically controversial) study carried out at Princeton University by Richard Gott and colleagues (Astron. J. 549 1) emphasized that confidence in these kinds of surprising results depends largely on having enough statistics to make a robust measurement, and asserted that the SCP team's data had accomplished this goal. Each of the two discovery papers — the 1998 paper by Riess et al. and the 1999 paper by Perlmutter et al. — has received over 3000 citations to date.

Within two years of the discovery, other experiments eliminated the possibility that had been mentioned in the SCP team's January 1998 poster of a combination of surprisingly curved space, low mass and a conspiracy of systematic biases that might keep open the possibility of no cosmological constant: cosmic-microwave-background experiments showed that space is not tightly curved, while measurements of galaxy clusters implied that the universe's mass is not low.

Ever since, the work has continued to be showered with prestigious prizes. Perlmutter won the E O Lawrence Award in 2002, the Feltrinelli Prize in 2005 and shared the Shaw Prize in 2006 with Schmidt and Riess. The 2003 Warner Prize and the 2004 Sackler Prize went to Riess, while the 2005 Padua Prize was shared by Perlmutter and Schmidt. The 2007 Gruber Cosmology Prize, worth a total of $500,000, was awarded a quarter each to Perlmutter and Schmidt, and a quarter each to the rest of the members of each team.

Since then, the leaders of the SCP and High-Z teams have worked together amiably and smoothly. Perlmutter and Schmidt published a review article together on the field in 2003 (arXiv:astro-ph/0303428v1), and Perl mutter and Riess also did so in 1999. At the Gruber prize award event, Perlmutter and Schmidt gave an amusing and unusual joint acceptance talk, alternating sentences every slide or so, handing the laser pointer back and forth as if it were a baton in a relay race. Occasionally, individuals on the High-Z team have pressed priority claims based on the timestamp of the final publications. In his book The Extravagant Universe, Kirshner wrote that the High-Z team benefited from the buzz the SCP team was making about the cosmological constant, but contentiously entitles two of his chapters "Getting it first" and "Getting it right". Since then, the leaders and key members of the collaborations — including Goldhaber, Kirshner and Perlmutter — have exchanged letters and e-mails arguing about the chronology and meaning of several of the events.

The critical point

Laws are like sausages, runs the old saw, for the less you know about how they are made, the more you respect the product — and one is sometimes tempted to say the same of scientific discoveries. But the saying is cleverer than true. It depends on one's expectations; those who truly understand human creativity have no difficulty with the idea that an objective effect can arise out of a messy human practice.

Science arises out of a much richer, looser and deeper ground than textbook images of its process suggest. Competition is not the only path to discovery, but competitive stories such as this one exhibit the ambiguous features of this process — regarding the nature of discovery, announcements and credit — in the way that other stories do not. They suggest that seemingly irrelevant details, such as who said what to whom and when, can strongly affect how individuals and teams conduct searches, make discoveries, and announce and publish results.

Competition stories can suggest, furthermore, that it is not always clear whether a discovery should be tethered to the date when a particular paper is submitted, to when it is announced to the public for the first time, or to when it is announced to the public in full without any caveats. Should a discovery be the date at which a clear presentation is made to the science community or perhaps to a specialist group meeting? Or should a discovery be tied to the date when the first evidence appears in someone's lab notebooks?

Such stories also suggest ambiguities in the nature of announcements, which differ depending on whether one is speaking to one's colleagues, the wider scientific community or the media. Does the race go to those who offer evidence for a conclusion and possible caveats, or to those who are unequivocal? These stories also suggest that it can be difficult to award credit when teams interact.

This story, too, nails the image of a scientific discovery as something that emerges all at once, with clarity and completeness, able to be announced with finality and certainty, as overly romantic and obsolete. Discoveries that emerge slowly from statistics do not make for good headlines, but are probably the wave of the future.

When and if a Nobel prize is eventually awarded to this amazing discovery, the Nobel committee will have to exercise Solomonic judgment in deciding how to apportion the award, to be shared maximally by three individuals, among two teams. For much of the small community helped one team or the other — or both — and the effect of the ambiguities, as most members of the teams are happy to see, is to spread around recognition. The final twist in this dramatic story of fierce competition may be the benevolent irony that the credit ends up going to both teams — in effect, to the entire supernova community.