Richard Barnes

University of California, Berkeley

Richard Barnes entered the University of Minnesota as an undergraduate philosophy major, but he also had a technical bent grounded in his early experiences with computers, so he took a physics course. One day, the professor broke his chalk while writing on the board.

“He looked down at it sadly,” says Barnes, a Department of Energy Computational Science Graduate Fellowship (DOE CSGF) recipient. Then he looked up and said: “Why would anyone ever be anything other than a physicist? It gives you a deep understanding of systems. Specialize in grad school.”

“Being undecided but technically oriented I was like, yes, that makes sense,” Barnes says. He later earned dual bachelor’s degrees, a year apart, in physics and philosophy.

Barnes’ humanities background informs his doctoral studies in computational ecology and geoscience at the University of California, Berkeley, under John Harte. He’s tackled diverse projects like developing perennial grain-producing plants that require fewer inputs, tracking how landscapes change over time and studying election gerrymandering.

His overarching goal: to apply computational science where it’s rarely been used.

“I’m trying to develop the fundamental algorithms that will allow us to scale” ecological and geospatial simulations to an unprecedented degree. Many such models today are either too removed or too imprecise “to be able to separate competing theories of how such processes unfold.”

In one paper, Barnes and colleague Adam Clark, then at the University of Minnesota, studied the distribution of salamanders across the Appalachian Mountains. It addressed a conundrum in evolutionary ecology, Barnes says: how species disperse over the ages. Any possible experiment takes so long “that it’s hard to say anything about what has happened to cause a pattern.” That leads to multiple explanations for what we see today. “The question is how do you choose between them?”

Researchers know salamanders are sensitive to moisture and temperature in their immediate surroundings but discount the influence of geographic and climate change. Because the Appalachians, now a low mountain range, once were as high as the Himalayas and the climate of the area was once far different from today, Barnes and Clark wanted to know how such drastic changes affected where salamanders are found in the present.

The researchers’ model ran evolution forward from 65 million years ago under changing or unchanging geography and climate parameters. They found that “the world in which you’re incorporating both the mountains collapsing and the climate changing tends to lead to distributions (of salamander populations) that better reflect those observed today” than models that omit those factors.

Barnes has used several high-performance computing (HPC) systems for his research, including Titan, a Cray XK7 at Oak Ridge National Laboratory, and Summitdev, a smaller test version of Summit, the lab’s newest machine and now the world’s fastest.

He’s also used Titan and Cori, a Cray XC40 at the National Energy Research Scientific Computing Center, in a recent collaboration with Lawrence Berkeley National Laboratory staff researching genome assembly on exascale computers, the next generation of machines due to arrive early in the next decade.

To rapidly decode the order of billions of base pairs that comprise DNA, researchers break the strands into millions of chunks, sequence each and then reassemble them in the correct order. Berkeley Lab researchers have created UPC, a programming framework that treats a supercomputer’s entire memory as accessible from any processor node to speed assembly.

Over summer 2018, Barnes examined whether graphics processing units (GPUs), low-power accelerators now used in many HPC systems, could do assembly effectively. He also began testing ways to leverage GPUs to align DNA segments.

“It can assemble the whole human genome in something like eight minutes,” Barnes says. “I talk to geneticists and they just kind of gape at me because this takes days, usually.”

The collaboration, which includes Berkeley Lab researchers Katherine Yelick, Erich Strohmaier and Aydın Buluç, is continuing and may be the subject for Barnes’ computer science master’s thesis.

Barnes’ DOE CSGF stint ends in 2019. He’ll continue his research as a Berkeley Institute for Data Science Fellow. He’s already exploring possible postdoctoral or faculty appointments.