Description
To study the dynamic behaviour of large chemical systems, its interactions are often parameterized with classical force fields. Traditionally, a force field is a sum of atomic two, three or four body-interactions. Polarizable force fields make better approximations of the interatomic interactions by explicitly accounting for electronic polarization. [e.g. see Verstraelen et al. J. Chem. Phys. 138, 074108 (2013)] To develop such models, one must go beyond the basic dipole polarizability and break down the linear response in local atomic contributions [Krishtal et al., J. Chem. Phys. 125, 034312 (2006)]. Such level of detail is needed to describe how two nearby molecular fragments mutually polarize each other, leading to the additional attractive forces that are of interest for force-field simulations.
In this work, we want to assess this partitioning by comparing it with the exact static polarizability for several well-chosen points along the groundstate dissociation curves of homonuclear first-row dimers: Li2, Be2, B2, C2, N2, O2 and F2. The total and partitioned static dipole polarizabilities can be obtained with the finite-difference method, by means of several ground-state calculations [Wouters et al., J. Chem. Phys. 136, 134110 (2012)]. The partitioning will provide additional insights in how to model the quantum mechanical contribution to linear response in polarizable and reactive force fields. In order to obtain eliable results along the entire dissociation path, a high-level electronic structure method is needed.
An exact electronic ground-state calculation corresponds to finding the smallest algebraic eigenvalue of the corresponding Hamiltonian, a large symmetric matrix. The linear dimension of this matrix grows exponentially with the number of single-particle degrees of freedom, which renders the exact solution for all ractically interesting cases infeasible. Fortunately, certain low-rank approximations of the exact solution converge quickly to Rules and regulations regarding applications to use the Flemish Supercomputer the exact solution with increasing rank. One such low-rank approximation is the matrix product state (MPS), which can be optimized by means of the density matrix renormalization group (DMRG). By exploiting the symmetry group of the Hamiltonian, an additional significant lowering of the computational cost is achieved. We have our own high-performance implementation of DMRG for quantum chemistry, heMPS2, which was released recently [Wouters et al., Comput. Phys. Commun. 185, 1501 (2014)].
