1: \begin{abstract}

2: 

3: A zero-site density matrix renormalization algorithm (DMRG0) is proposed

4: to minimize the energy of matrix product states (MPS). Instead of

5: the site tensors themselves, we propose to optimize sequentially the ``message'' tensors between neighbor sites, which contain the singular values of the bipartition. This

6: leads to a local minimization step that is independent of the physical dimension of the site. Conceptually, it separates the optimization and decimation steps in DMRG. Furthermore, we introduce two new global perturbations

7: based on the optimal low-rank correction to the current state, which are used to avoid local minima. They are determined variationally as the MPS closest to

8: the one-step correction of the Lanczos or Jacobi-Davidson eigensolver,

9: respectively. These perturbations mainly decrease the

10: energy and are free of hand-tuned parameters. Compared to existing single-site enrichment proposals, our approach

11: gives

12: similar convergence ratios per sweep while the computations are cheaper

13: by construction. 

14: Our methods may be useful in systems with many physical degrees of freedom per lattice site. We test our approach on the periodic Heisenberg

15: spin chain for various spins, and on free electrons on the lattice.

16: 

17: 

18: 

19: \end{abstract}

20: