Bleary eyes and a taxed brain are common at Materials Research Society (MRS) Meetings, but for the first time, a group of intrepid attendees had a very different reason for their weariness than the typical gauntlet of talks and networking sessions. During the 2014 MRS Fall Meeting in Boston, 14 materials scientists came together for 24 hours for MatHack, the world’s first materials hackathon, to solve real materials problems. A hackathon is a sprint computer programming competition where participants collaborate to create software from scratch in intense sessions over one or two days. Sponsored, in part, by Citrine Informatics and driven by two of its founders (Bryce Meredig and Greg Mulholland) in collaboration with the MRS Academic Affairs Committee, MatHack participants pitched ideas, formed teams, spent one night writing code, and presented their work to a panel of judges from across the materials community.

The idea behind a hackathon is to very quickly build functional (yet imperfect) software to lay the foundation for further development in the future. Such events are common in Silicon Valley; Google and Facebook are renowned hackathon hosts and sponsors. Mark Zuckerberg, founder and chief executive officer of Facebook, has explained that “hacking just means building something quickly or testing the boundaries of what can be done.”^{Footnote 1} The participants at the first MatHack did just that, demonstrating that even in a field like materials science—traditionally associated with longer-term laboratory investigations—people can produce creative, meaningful scientific software contributions in a very short time.

MatHack participants hailed from wide-ranging backgrounds within the materials community. Some were experienced computational materials scientists with tremendous coding backgrounds. Others were experimentalists with practical problems to solve as well as professors and postdocs with a desire to build something new. The participants included people from universities all over the world, spanning academia, industry, and national labs.

Hackathon participants, judges, and organizers (front row, left to right): Brendan Nagle (Dartmouth College), Katie Van Aken (Drexel University), Sabrina Ball (MIT), Ioan-Bogdan Magdau (The University of Edinburgh); (second row, left to right): Anubhav Jain (Lawrence Berkeley Lab), Timothy Large (Microsoft Applied Sciences), Nicole Adelstein (Lawrence Livermore National Lab), Rick Barto (Lockheed Martin), Susan Ermer (Lockheed Martin), Bryce Meredig (Citrine Informatics), Guoqiang Xu (MIT); (third row, left to right): Andre Schleife (Univ. Illinois at Urbana-Champaign), Oleg Rubel (McMaster/TBRRI), Wenhao Sun (MIT), Greg Mulholland (Citrine Informatics).

Participants first gathered in The Hub, the central area of the Fall Meeting, and gave 30-second pitches for ideas that addressed a very broad prompt: “Build a piece of software that would be useful to a materials scientist.” Some pitches were precise answers to lab pains that had simmered for years, while others were more general, outlining a rough idea of what a team might build together. While some projects could expand into entire doctoral dissertation topics, others tried to solve everyday problems that present themselves when working with materials, or proposed new means of outreach to inspire the next generation of diverse, engaged materials scientists. Ideas ranged from a tool to communicate the status of characterization tools within a lab, to automatically solving complex equivalent circuits, to density functional theory (DFT) approaches to quasicrystals, to crystal visualization.

After the pitches were complete, each presenter set up at a table, and for the next 30 minutes, participants met and talked with one another to learn more about the vision, technical challenges, and people involved with each idea. Then the teams were formed around many of these ideas. Most participants had never met before and built relationships over the course of the 24 hours.

Once teams formed, the real work began. While everyone had selected hard problems to solve, they did not all have the programming expertise or requisite knowledge to build the right systems. In some cases, people had an idea of how to execute from the outset; in other cases, they had to learn an entirely new programming language. The common thread, though, was a willingness and desire to learn new things, ask for help, and share ideas liberally. In typical hackathon fashion, sleep was scarce. Gallons of coffee and ice cream and pounds of nuts and other snacks fueled the teams through the night.

The next morning, participants presented their ideas to the panel of judges. This diverse group of judges then selected the top three teams for cash prizes. Though most of the faces were tired, no one appeared ready to stop working. Bruce Clemens, Stanford professor and former MRS President, said it best when he declared, “These teams blew me away!”

Hackathons often have themes, ranging from health care to civic engagement to education. MatHack was the first time that a hackathon’s theme was materials science. Some projects will likely continue; others may not. What will certainly grow is the influence of software and hackathon thinking in materials at large. What can we accomplish if we boil materials challenges down to their bare essence and ask smart teams to focus intensely on them?

The collaborations and relationships that formed at MatHack, over a long night bathed in the glow of laptop screens, will certainly continue throughout the participants’ scientific careers. This is the reason the MRS Meetings exist—to build such communities—and MatHack fits right into this mold.

WINNERS

First Place: Phone-on Flow

Nicole Adelstein and Andre Schleife

Team Phone-on Flow built a functioning three-dimensional crystal viewer. Using an Android phone, a student could visit a special web page and (using Google Cardboard,^{Footnote 2} take an immersive virtual tour through a crystal structure. Even to the judging panel, who has seen their share of crystals, the visualization was inspiring. As an outreach tool, such an inexpensive and interactive project could open up materials to a new generation of students.

Second Place: MatHack QuasiCrystal

Wenhao Sun

Sun developed a new way to use DFT to simulate quasicrystals, which can be thought of as the material equivalent of the number π: perfectly ordered but eschewing a repeating pattern. More importantly, they are a class of materials that shows promise for many applications but resists simple treatment with atomistic simulations. MatHack QuasiCrystal is the first step in being able to simulate these materials to understand them better.

Third Place: Directed Materials Design (DMD)

Andrew Long and Ioan-Bogdan Magdau

Team DMD took an operations research approach to understanding materials. By looking at large-scale data sets, they built machine learning models to rapidly predict how various perovskite materials would work as water-splitting materials, without requiring expensive supercomputing time or the upfront investment of experiments.

TEAMS

CIVR

Sabrina Ball, Brendan Nagle, and Katie Van Aken

Team CIVR started with a clear question: “How can I automate the solving of equivalent RC circuits for electrochemical systems?” For the experimentalist, this is a hard problem. It involves solving huge sets of equations and fitting them to experimentally measured I–V curves.

MacHack

Guoqiang Xu

MacHack is a tool to analyze the grain structure of a material using optical microscope images. It has a different take from most approaches, because it uses multiple images of the same spot under different lighting conditions, combined with machine learning to identify grain boundaries.

MatMod

Gabriela Correa and Oleg Rubel

Team MatMod wanted to address the problem of DFT (see text earlier) being computationally expensive. This is no small feat—Walter Kohn and John Pople shared the Nobel Prize in Chemistry in 1998 for developing the method and operationalizing it in working software. MatMod’s approach was to use alternative k-point sampling techniques to reduce the complexity of DFT calculations for a particular system without substantially reducing accuracy.