CarlosPinho
Membro
What is [SIZE=+1]E[/SIZE][SIZE=+2]O[/SIZE][SIZE=+1]N[/SIZE]?
A common problem in theoretical chemistry, condensed matter physics and materials science is the calculation of the time evolution of an atomic scale system where, for example, chemical reactions and/or diffusion occur. Generally the events of interest are quite rare (many orders of magnitude slower than the vibrational movements of the atoms), and therefore direct simulations, tracking every movement of the atoms, would take thousands of years of computer calculations on the fastest present day computer before a single event of interest can be expected to occur, hence the name [SIZE=-1]E[/SIZE]O[SIZE=-1]N[/SIZE], which is an immeasurable period of time.
The Henkelman Research Group is interested in calculating the long time dynamics of systems. We have developed a method for doing this through distributed computing where a server sends out small data packets for calculation to clients, e.g. over the internet. So, instead of the entire calculation being done on a single processor, it is done on many client computers worldwide. After finishing it's calculation, each client computer sends it's results back to the server, which summarizes the results and sends out more jobs.
Scientific Overview
The Henkelman Research Group is interested in calculating the long time dynamics of systems. We have developed a method for doing this, using the dimer method for saddle point finding combined with the kinetic Monte Carlo to advance the system over barriers. This method has been used to study island formation and growth. An aluminum system was used to develop a serial implementation of our methods. This project will be devoted to ice growth.
In broad strokes, a system is advanced along a series of minimum energy configurations which are all separated from one another by one first order saddle point. From each new minimum structure, a set of independent dimer searches is launched. The dimers crawl up the potential basin away from the minimum and converge upon saddle points. After a sufficient number of saddles are found, the searches are stopped, and one is chosen. This is done by weighted each possible process (defined by a saddle point) by the rate for that process. The rate of a process is based upon the energy barrier required to cross the corresponding saddle point, and a harmonic prefactor. The system is advanced over the chosen process and the process is repeated.
Our method is particularly suited to distributed computing. The swarm of dimer searches launched from each minimum (from the server) are independent and can be run on separate machines (clients). Furthermore, only a tiny amount of data is needed to initiate each search on the clients and report the results back to the server. The small communication requirements means that the system can be implemented over the internet. Finally, it is not critical to get results back from searches. If a search goes bad, or a client does not report a results, the server can simply go on without it, or farm more searches out to different machine.
Related articles (from old eon project)
How to join eOn
New eon2: http://eon.ices.utexas.edu/eon2/index.php
Portugal@Home: http://eon.ices.utexas.edu/eon2/team_display.php?teamid=818
Old eon: http://eon.cm.utexas.edu/index.php
A common problem in theoretical chemistry, condensed matter physics and materials science is the calculation of the time evolution of an atomic scale system where, for example, chemical reactions and/or diffusion occur. Generally the events of interest are quite rare (many orders of magnitude slower than the vibrational movements of the atoms), and therefore direct simulations, tracking every movement of the atoms, would take thousands of years of computer calculations on the fastest present day computer before a single event of interest can be expected to occur, hence the name [SIZE=-1]E[/SIZE]O[SIZE=-1]N[/SIZE], which is an immeasurable period of time.
The Henkelman Research Group is interested in calculating the long time dynamics of systems. We have developed a method for doing this through distributed computing where a server sends out small data packets for calculation to clients, e.g. over the internet. So, instead of the entire calculation being done on a single processor, it is done on many client computers worldwide. After finishing it's calculation, each client computer sends it's results back to the server, which summarizes the results and sends out more jobs.
Scientific Overview
The Henkelman Research Group is interested in calculating the long time dynamics of systems. We have developed a method for doing this, using the dimer method for saddle point finding combined with the kinetic Monte Carlo to advance the system over barriers. This method has been used to study island formation and growth. An aluminum system was used to develop a serial implementation of our methods. This project will be devoted to ice growth.
In broad strokes, a system is advanced along a series of minimum energy configurations which are all separated from one another by one first order saddle point. From each new minimum structure, a set of independent dimer searches is launched. The dimers crawl up the potential basin away from the minimum and converge upon saddle points. After a sufficient number of saddles are found, the searches are stopped, and one is chosen. This is done by weighted each possible process (defined by a saddle point) by the rate for that process. The rate of a process is based upon the energy barrier required to cross the corresponding saddle point, and a harmonic prefactor. The system is advanced over the chosen process and the process is repeated.
Our method is particularly suited to distributed computing. The swarm of dimer searches launched from each minimum (from the server) are independent and can be run on separate machines (clients). Furthermore, only a tiny amount of data is needed to initiate each search on the clients and report the results back to the server. The small communication requirements means that the system can be implemented over the internet. Finally, it is not critical to get results back from searches. If a search goes bad, or a client does not report a results, the server can simply go on without it, or farm more searches out to different machine.
Related articles (from old eon project)
- Xu Lijun and Graeme Henkelman Adaptive kinetic Monte Carlo for first-principles accelerated dynamics, J. Phys. Chem. 129, 114104, (2008).
- Graeme Henkelman and Hannes Jónsson Multiple time scale simulations of metal crystal growth reveal importance of multi-atom surface processes, Phys. Rev. Lett. 90, 116101, (2003).
- Graeme Henkelman and Hannes Jónsson Long time scale kinetic Monte Carlo simulations without lattice approximation and predefined event table, J.Chem.Phys. 115, 9657, (2001).
- Graeme Henkelman and Hannes Jónsson A dimer method for finding saddle points on high dimensional potential surfaces using only first derivatives, J.Chem.Phys. 111, 7010, (1999).
How to join eOn
- This project uses BOINC. If you're already running BOINC, select Attach to Project. If not, download BOINC.
- When prompted, enter http://eon.ices.utexas.edu/eon2/
- If you're running a command-line or pre-5.0 version of BOINC, create an account first.
- If you have any problems, get help here.
New eon2: http://eon.ices.utexas.edu/eon2/index.php
Portugal@Home: http://eon.ices.utexas.edu/eon2/team_display.php?teamid=818
Old eon: http://eon.cm.utexas.edu/index.php
Última edição: