Research Topics of the Behler Group
Lehrstuhl für Theoretische Chemie, Ruhr-Universität Bochum
Direct link to the
list of NN potentials.
The main research topic of the Emmy Noether group is the development and application of efficient interatomic potentials based on artificial Neural Networks. In contrast to conventional empirical potentials, which have a fixed functional form based on a number of assumptions and approximations, Neural Networks represent a purely mathematical approach with high numerical accuracy. The correct physical description of the systems under investigation is introduced via a large number of accurate electronic structure reference calculations. Based on a large number of these ab initio energies and forces for different atomic configurations, a continuous representation of the potential-energy surface is obtained by a Neural Network interpolation, i.e. the Neural Network 'learns' the shape of the potential-energy surface. Neural Networks are very flexible and, once properly trained, provide an accurate analytic expression for the potential-energy surface.
1. Development of Neural Network Potentials
Neural Network potentials have a number of advantages:
However, the flexible functional form of Neural Networks results in some disadvantages as well:
high numerical accuracy in reproducing the total energies of the reference ab initio calculations
no physical approximations needed
no 'ad hoc' introduction of arbitrary energy terms necessary
analytic gradients for the calculation of forces and the stress tensor are available
evaluation of Neural Network energies and forces is very fast (about 100 atoms per core per second), typically 4-5 orders of magnitude faster than DFT at the GGA level
in contrast to force fields, no classification of atoms and/or bonds is needed, Neural Network potentials are intrinsically "reactive"
input has the same form as for electronic structure calculations: atomic positions and nuclear charges
all types of bonding are treated in the same way (covalent, metallic, vdW etc.), no system-dependent adaptions of the functional form is required
any electronic structure method can be used as reference method, a coupling to many quantum chemistry codes is easily possible
very efficient parallelization
because of the high flexibility a large number of reference calculations is needed
formally no extrapolation capabilities (a potential made just for bulk will not work for clusters, but of course potentials can be trained to both)
the accuracy of the Neural Network potential is limited by the accuracy of the underlying electronic structure method
currently, the number of different chemical elements is limited to 3-4, but the number of atoms of each element can be very large
Because of these advantages and disadvantages, Neural Network potentials are most useful to speed up molecular dynamics simulations, if complex bonding situations or chemical reactions are present which cannot be described accurately by conventional potentials. We believe that in such cases Neural Network potentials are the method of choice to perform long molecular dynamics simulations of large systems, which are not directly accessible by ab initio MD.
We are constantly working on the further development of the Neural Network methodology to improve the accuracy and performance of these potentials.
In order to construct Neural Network potentials for a variety of systems, we have written our own Neural Network package RuNNer.
We are constructing Neural Networks for a variety of systems in the fields of Materials Science and Surface Science.
The first successful applications refer to the study of phase diagrams of several solid systems, like the high-pressure phase diagram of silicon, the phase diagram of sodium, and the graphite-diamond coexistence line of carbon. Recently, we have started to study also molecular systems and liquids.
2. Applications of Neural Network Potentials
For further details, please have a look at our
list of applications.
Using well-established theoretical surface science techniques like density-functional theory and 'ab initio thermodynamics', we study the stability of clusters at metal surfaces (SFB558), the structure and formation of self-assembled monolayers of organic molecules at metal surfaces, as well as dynamical processes at the gas/solid interface, like e.g. the dissociation of oxygen molecules at metal surfaces.
3. Surface Science
4. Solvation Science
In the context of the research focus Solvation Science@RUB
we have started to study the properties of water employing high-dimensional NN potentials.
These potentials allow to describe the reactivity and dissociation of fully flexible water molecules with about DFT accuracy. Since the functional form of NNs is very flexible and the resulting potentials are not bound to any particular type of bonding, they can easily be extended to describe the interaction between water and other molecules, or even between water and solid surfaces. Therefore, NN potentials represent an ideal tool to study solvation in a variety of systems, from bulk solutions to chemical processes at interfaces. Our current research focusses on the development of NN-based potentials to study these systems.