The RuNNer Code
In order to develop Neural Network potential-energy surfaces for a variety of system, we have developed our own NN code called RuNNer. The code is currently still under development and not yet publicly available.
Lehrstuhl für Theoretische Chemie, Ruhr-Universität Bochum
RuNNer has the following features:
unlimited number of degrees of freedom (atoms)
training data can be obtained from arbitrary electronic structure methods and codes
training using energies and forces
periodic and non-periodic system
several types of symmetry functions are available
several types of activation functions are available
arbitrary topology of the atomic neural networks
provides energies and analytic derivatives (forces and stress tensor)
The methodology of RuNNer is published in the following papers:
J. Behler, and M. Parrinello, Phys. Rev. Lett. 98, 146401 (2007).
J. Behler, R. Martonak, D. Donadio, and M. Parrinello, phys. stat. sol. (b) 245, 2618 (2008).
J. Behler, J. Chem. Phys. 134, 074106 (2011).
More details will follow soon.
If you are interested in learning more about RuNNer, please feel free to
Link to the internal RuNNer page