A simple energy argument indicates that the existence of an eigenvalue of order N−1 in the second-order reduced density matrix, or 2-matrix, implies macroscopic correlation characterized by the corresponding eigengeminal.
Usually, variational calculations for the second-order reduced density matrix are performed subject to the constraint that the ``P'' and ``Q'' matrices are positive semidefinite, which only constrains the lowest eigenvalue of these matrices. We characterize the highest eigenvalue of these matrices and discuss how the associated constraint (which is related to the ground-state energy of the Hamiltonians $H=\ensuremath{-}P$ and $H=\ensuremath{-}Q$) can be implemented in practical calculations. This necessary condition for N-representability should help ensure that the second-order reduced density matrix is not ``overcorrelated.''
The expectation value of the Hamiltonian H for a many-electron system may be expressed in terms of the reduced second-order density matrix and the “reduced” Hamiltonian K for a two-electron system. It may be shown that the antisymmetric component HAS of the original Hamiltonian is an “outer projection” of K with respect to the antisymmetrizer O, so that HAS = OKO. This result implies that the eigenvalues of K are, in order, lower bounds to the eigenvalues of HAS, i.e. to the eigenvalues of H associated with the antisymmetry requirement. The expectation values of K are hence often below the ground state energy of H, and the eigenvalues may be used for the calculation of lower bounds. Some numerical applications to atoms serve as an illustration, and it is shown that the lower bounds become worse as the number of electrons increases. The implications for the “representability problem” are discussed.
We present stochastic diagonalization results for the ground-state energy and the largest eigenvalue of the two-fermion density matrix of the BCS reduced Hamiltonian, the Hubbard model, and the Hubbard model with correlated hopping. The system-size dependence of this eigenvalue is used to study the existence of Off-Diagonal Long-Range Order in these models. We show that the model with correlated hopping and repulsive on-site interaction can exhibit Off-Diagonal Long-Range Order. Analytical results for some special limiting cases indicate that Off-Diagonal Long-Range Order not always implies superconductivity.