Abstract
Four common pure fluids were chosen to elucidate the reliability of reactive force fields in estimating bulk properties of selected molecular systems: CH4, H2O, CO2 and H2. The pure fluids are not expected to undergo chemical reactions at the conditions chosen for these simulations. The ‘combustion’ ReaxFF was chosen as reactive force field. In the case of water, we also considered the ‘aqueous’ ReaxFF model. The results were compared to data obtained implementing popular classic force fields. In the gas phase, it was found that simulations conducted using the ‘combustion’ ReaxFF formalism yield structural properties in reasonable good agreement with classic simulations for CO2 and H2, but not for CH4 and H2O. In the liquid phase, ‘combustion’ ReaxFF simulations reproduce reasonably well the structure obtained from classic simulations for CH4, degrade for CO2 and H2, and are rather poor for H2O. In the gas phase, the simulation results are compared to experimental second virial coefficient data. The ‘combustion’ ReaxFF simulations yield second virial coefficients that are not sufficiently negative for CH4 and CO2, and slightly too negative for H2. The ‘combustion’ ReaxFF parameterisation induces too strong an effective attraction between water molecules, while the ‘aqueous’ ReaxFF yields a second virial coefficient that is in reasonable agreement with experiments. The ‘combustion’ ReaxFF parameterisation yields acceptable self-diffusion coefficients for gas-phase properties of CH4, CO2 and H2. In the liquid phase, the results are good for CO2, while the self-diffusion coefficient predicted for liquid CH4 is slower, and that predicted for liquid H2 is about nine times faster than those expected based on classic simulations. The ‘aqueous’ ReaxFF parameterisation yields good results for both the structure and the diffusion of both liquid and vapour water.
Acknowledgments
We acknowledge the financial support from the U.S. Department of Energy, Office of Basic Energy Sciences, under Contract No. DE-SC0006878 (Division of Chemical Sciences, Geosciences, and Biosciences), Geosciences Program. Additional financial support was provided by the A. P. Sloan Foundation via the Deep Carbon Observatory administered by the Carnegie Institution for Science. AS acknowledges financial support from the European Union via the Marie Curie Career Integration Grant No. 2013-CIG-631435. Generous allocations of computing time were provided by the National Energy Research Scientific Computing Center (NERSC) at Lawrence Berkeley National Laboratory, Berkeley, CA. NERSC is supported by the DOE Office of Science. We are also grateful to the Oklahoma Supercomputer Center for Education and Research (OSCER), for access to high-performance computing. We acknowledge the suggestions from the anonymous reviewers, who, among other important comments, reminded us that experimental second virial coefficients could be used to compare against simulation results and that the simulated self-diffusion coefficients strongly depend on the simulation box size.