Brenhin Keller

Princeton University

When he started his graduate studies in 2011, Brenhin Keller had planned on a laboratory-based research career focusing on the chemistry and features of ancient rocks. But at the time, his Princeton University Ph.D. advisor, Blair Schoene, was still setting up equipment in his new laboratory. He and Keller instead chose a computational project Keller could work on immediately, until the lab was ready.

That research – analyzing geochemical databases to probe fundamental questions about the Earth’s early geology – soon led to a paper for the prestigious journal Nature. By the time it came out, Keller also had recognized that computation let him address questions other geochemists hadn’t been able to examine quantitatively. With the support of a Department of Energy Computational Science Graduate Fellowship (DOE CSGF), this side project grew into Keller’s primary research focus.

As a Cornell University undergraduate, Keller noticed there were areas of geology and geochemistry where computation could be useful but hadn’t been widely applied. “Even though I didn’t get into computation much as an undergrad, it was in the back of my mind,” he says. He arrived at Princeton with little programming experience but got up to speed in courses and in consultations with colleagues whose work often used computational techniques.

Schoene had experience working with rocks from the Archaean Eon, between 4 billion and 2.5 billion years ago. At the end of the eon oxygen first became abundant in the atmosphere, dramatically altering how chemicals like sulfur and iron cycle from soil to air to organisms and back again and leading to one of Earth's earliest ice ages.

“There are a lot of qualitative observations about the Archaean to suggest that there may have been a quantitative difference in the way things worked back then,” Keller says. But those earlier studies were based on the analysis of only a few hundred to a couple of thousand rocks. Within the last 20 years, geochemists have compiled data about far more rock samples into online databases such as EarthChem.

Keller and Schoene realized computational tools could mine this chemical information so they could model magma formation during the early Archaean. “Instead of having 1,000 or 2,000 samples, which is what you’d typically see in a compilation paper at the time, we had 70,000,” Keller says.


To scrutinize these data sets, Keller used weighted bootstrap resampling, a statistical technique that ensured the analysis accurately represented the entire globe and the uncertainties within the field measurements.

For example, the researchers had to account for samples’ proximity to each other. “If there’s one area where we sampled a whole bunch of rocks right next to each other,” Keller says, “each one of them doesn’t contain as much new information as a (sample) from someplace far away that’s the only one of its kind.” There’s also some uncertainty inherent in analyses of a rock’s age and chemistry. “If you look at rocks of any given age, the variability of their composition at any one time is much larger than their variability over time.” Keller made that uncertainty part of the resampling process.

The researchers’ models, as detailed in their 2012 paper, showed long-term continuous cooling of the Earth’s mantle over the Archaean Eon. But around 2.5 billion years ago, the cooling patterns abruptly steepened. At that same time the continental rock record shows rapid changes in the abundances of a range of trace elements. Taken together, these suggest a modification in the process of crustal differentiation, in which mafic (high in magnesium, low in silicon) magma derived from Earth's mantle evolves into less-dense felsic (high-silicon) magma.

That timing also is consistent with when an abundance of oxygen first appeared in Earth’s atmosphere, a critical point in life’s evolution. Although this oxygen ultimately comes from photosynthesis, some of the details of this record don't align. Fossil evidence suggests that oxygen-producing photosynthesizers evolved well before 2.5 billion years ago, but the gas didn’t accumulate then, indicating that it was consumed faster than it was produced.

Keller’s analysis suggests an explanation based on chemical changes in rocks dating from the Archaean. They indicate that crustal differentiation occurred at high pressure deep within Earth, which is one way to produce magmas (and associated volcanic gases) that have more electrons available for bonding to other elements. Such magmas and volcanic gases would consume oxygen from the atmosphere, reducing it until the end of the Archaean.

Keller plans to incorporate lab and field research into future projects, but computational work will remain his core focus.  “This project and the DOE CSGF have turned me into more of a computational scientist than I ever thought I would be,” he says.

Read the entire article in DEIXIS, the DOE CSGF annual. [PDF, pages 7-8]

Image caption: An estimated sample density map reveals the persistence through geologic time of silicate rock composition with abundant basaltic (about 50 percent silica) and granitic (about 70 percent silica) magmas but little in between. Credit: Brenhin Keller.