How well do Car-Parrinello simulations reproduce the Born-Oppenheimer surface? Theory and examples

We derive an analytic expression for the average difference between the forces on the ions in a Car-Parrinello simulation and the forces obtained at the same ionic positions when the electrons are at their ground state. We show that for common values of the fictitious electron mass, a systematic bias may affect the Car-Parrinello forces in systems where the electron-ion coupling is large. We show that in the limit where the electronic orbitals are rigidly dragged by the ions the difference between the two dynamics amounts to a rescaling of the ionic masses, thereby leaving the thermodynamics intact. We study the examples of crystalline magnesium oxide and crystalline and molten silicon. We find that for crystalline silicon the errors are very small. For crystalline MgO the errors are very large but the dynamics can be quite well corrected within the rigid-ion model. We conclude that it is important to control the effect of the electron mass parameter on the quantities extracted from Car-Parrinello simulations. (C) 2002 American Institute of Physics.