eOn

eOn (styled eOn2 during its time on the BOINC platform) was a completed volunteer computing project of the Henkelman Research Group at the University of Texas at Austin, run out of the university's Institute for Computational Engineering and Sciences (ICES). The project applied theoretical chemistry methods, chiefly transition state theory combined with kinetic Monte Carlo, to model the evolution of atomic scale systems over timescales far longer than those reachable with conventional molecular dynamics. Its name is drawn from eon, a reference to the immeasurably long timescales the software was built to reach.

eOn
Project
Status	Completed
Category	Materials Science
Compute	CPU
Requires	None
Development
Developer	Henkelman Research Group, University of Texas at Austin, in collaboration with the Jónsson Research Group, University of Iceland
Author	Graeme Henkelman and Hannes Jónsson
Sponsor	Institute for Computational Engineering and Sciences (ICES), University of Texas at Austin
Maintainer	Discontinued as a BOINC project; the underlying EON software is maintained independently (see Legacy)
Initial release	September 7, 2010  (16 years ago)
Completed	Yes
Discontinued	May 26, 2014  (12 years ago)
Repository	https://github.com/TheochemUI/eOn
Software
Written in	C++ (client), Python (server)
Operating system	Windows, Linux, macOS
Size	Not documented
Metadata
Website	http://eon.ices.utexas.edu/eon2/
License	GPL-3.0

eOn joined the BOINC network in September 2010 and ran until it was retired in May 2014. The underlying EON research code, distinct from the retired volunteer computing project, continues to be developed and used within the computational chemistry community.

History

Origins

The EON software grew out of long-running work in the Henkelman group (University of Texas at Austin) and the Jónsson group (then at the University of Iceland) on methods for simulating rare-event dynamics in solids, building on the dimer method the two researchers had developed for locating saddle points on potential energy surfaces without needing a known final state. Before it was ported to BOINC, the group ran an earlier, home-built distributed computing system under the name eOn; forum members reported that the "old eOn client application" was subsequently rebuilt on BOINC's client/server architecture as eOn2, with the intention of later adding dynamics trajectories and ab initio (density functional theory) work units. For a time two separate project addresses existed, one on the group's older cm.utexas.edu domain and one on ices.utexas.edu; volunteers were directed to the ICES address as the project's active server.

Launch on BOINC

eOn was announced as a new BOINC project on 7 September 2010 by David P. Anderson, describing it as a study of "the dynamic simulation of physical and chemical processes over a time scale which is much longer than can be reached with traditional molecular dynamics." Because the generation of new work units depended on the results already returned by earlier ones, eOn could typically only keep a handful of tasks in flight on any one host at a time, a workflow characteristic of its adaptive kinetic Monte Carlo approach.

Early participants reported a bumpy rollout, including client crashes on Mac OS X and complaints about long-running, unchecked work units, alongside genuine enthusiasm from teams that had followed the Henkelman group's earlier, non-BOINC eOn effort.

The University of Texas at Austin, home to the Institute for Computational Engineering and Sciences, where eOn was developed and hosted.

Scientific method

eOn's simulations targeted rare event systems, ones in which the interesting physics, such as diffusion in solids or chemical reactions at surfaces, consists of infrequent transitions between long-lived stable states, separated by fast atomic vibrations that are many orders of magnitude quicker than the transitions themselves. A direct molecular dynamics simulation tracking every vibration would need to run for a computationally unreasonable length of time before a single event of interest occurred.

To get around this, EON combines transition state theory with kinetic Monte Carlo in an approach called adaptive kinetic Monte Carlo (aKMC). Rather than drawing transition rates from a predefined table of possible events, aKMC discovers the available transitions on the fly by searching for first-order saddle points around the current state. Under the harmonic approximation to transition state theory, the rate of an individual transition takes the Arrhenius-like form

$k=\nu e^{-\mathrm{\Delta }{E}_a/k_{B}T}$

where $\mathrm{\Delta }{E}_a$ is the energy barrier between the initial minimum and the saddle point, $T$ is the temperature, $k_B$ is the Boltzmann constant, and the prefactor $\nu$ is calculated from the vibrational modes at the minimum and the saddle point.

Saddle points themselves were located with the dimer method, in which two closely spaced replicas of the system ("the dimer") are used to estimate the direction of lowest curvature on the potential energy surface using only first derivatives of the energy, avoiding the cost of computing a full Hessian:

${\mathbf{𝐑}}_{1,2}={\mathbf{𝐑}}_0\pm \mathrm{\Delta }R\hat{N}$

The dimer is rotated to align with this minimum mode and then walked uphill along it until it converges on a saddle point. Besides aKMC, the EON software also implemented parallel replica dynamics, hyperdynamics, and global optimization methods including simulated annealing, basin hopping, and minima hopping.

Software architecture

A potential energy surface with a saddle point separating two minima, the kind of feature EON's dimer method searches for.

EON used a client/server design: computationally intensive force and energy evaluations ran on volunteer client machines, while a central server managed the overall state-to-state evolution of the simulation, for example driving the kinetic Monte Carlo bookkeeping in aKMC runs. The server was written in Python and the client in C++, the latter chosen so that a self-contained compiled executable could be distributed easily to volunteers. Beyond BOINC, the same codebase could also run on a single workstation, on cluster queuing systems, and on large parallel machines via MPI. The code was released under version 3 of the GNU General Public License.

Discontinuation

On 26 May 2014, the eOn team posted to the BOINC forums that the project was being retired from the BOINC network. Volunteer forums reported that the eOn server was taken offline shortly afterward, with one community post noting that it had been "retired" with little fanfare and that "the server has been switched off already." A follow-up post from the project team described the decision as a practical one: the project had been useful but had become increasingly cumbersome to maintain relative to the group's other computational resources.

Legacy

The retirement of the BOINC-facing eOn2 project did not end development of the underlying EON software. Around 2018, the codebase was relicensed from GPL-3.0 to the BSD 3-Clause license so that it could be incorporated into the Amsterdam Modeling Suite, a commercial computational chemistry platform. From 2019 onward, the code was migrated from its original Subversion repository to GitHub, and it continues to be maintained as an open-source research code, including a Gaussian process-accelerated dimer method contributed as part of doctoral work associated with the University of Iceland. This ongoing development is separate from the discontinued volunteer computing project and does not distribute work to BOINC volunteers.

Publications

Illustration of tilt and twist grain boundaries, the kind of defect studied in the 2009 EON simulation of a copper grain boundary.

The Berkeley BOINC project maintains a list of publications with results arising from BOINC-based computing, including work that made use of the eOn volunteer computing project:

1. Chill, Samuel T., Matthew Welborn, Rye Terrell, Liang Zhang, Jean-Claude Berthet, Andreas Pedersen, Hannes Jónsson, Graeme Henkelman. EON: software for long time simulations of atomic scale systems. Modelling and Simulation in Materials Science and Engineering 22(5), 055002 (2014). Template:DOI.
2. Pedersen, Andreas, Graeme Henkelman, Jakob Schiøtz, Hannes Jónsson. Long time scale simulation of a grain boundary in copper. New Journal of Physics 11(7), 073034 (2009). Template:DOI.
3. Mei, Donghai, Lijun Xu, Graeme Henkelman. Potential Energy Surface of Methanol Decomposition on Cu(110). The Journal of Physical Chemistry C (2009). Template:DOI.
4. Xu, Lijun, Donghai Mei, Graeme Henkelman. Adaptive kinetic Monte Carlo simulation of methanol decomposition on Cu(100). The Journal of Chemical Physics (2009). Template:DOI.
5. Xu, Lijun, Graeme Henkelman. Adaptive kinetic Monte Carlo for first-principles accelerated dynamics. The Journal of Chemical Physics (2008). Template:DOI.
6. Henkelman, Graeme, Hannes Jónsson. Multiple Time Scale Simulations of Metal Crystal Growth Reveal the Importance of Multiatom Surface Processes. Physical Review Letters (2003). Template:DOI.

No additional eOn-specific publications beyond these six were found in the BOINC publications registry at the time of writing.

See also

• BOINC projects
• Einstein@Home
• The Lattice Project

External links

• https://web.archive.org/http://eon.ices.utexas.edu/eon2/
• https://github.com/TheochemUI/eOn
• https://henkelmanlab.org/research/ltd/