| density matrix renormalization group | |
|---|---|
| Name | Density Matrix Renormalization Group |
| Invented by | Steven R. White |
| Year | 1992 |
| Field | Quantum many-body physics |
| Related | Matrix product state; Tensor network; Renormalization group |
density matrix renormalization group
The density matrix renormalization group (DMRG) is a numerical variational technique for obtaining approximate ground states and low-energy excitations of strongly correlated quantum many-body systems. It revolutionized the study of one-dimensional quantum spin chains and interacting fermion systems by providing high-precision results where traditional perturbation theory and exact diagonalization were intractable. DMRG matters in Quantum Physics because it bridged numerical methods and analytical concepts such as renormalization group and entanglement entropy.
DMRG was introduced by Steven R. White in 1992 as an improvement over real-space renormalization group methods pioneered by Kenneth G. Wilson. Early applications targeted the Heisenberg model and the Hubbard model on one-dimensional lattices, quickly demonstrating orders-of-magnitude better accuracy than previous truncation schemes. The method's success spurred development at institutions such as the Max Planck Institute for the Physics of Complex Systems, IBM Research, and university groups including University of California, Santa Barbara and Stanford University. DMRG established strong links between computational physics and quantum information theory after works by Guifre Vidal, Frank Verstraete, and Israel Affleck connected DMRG to matrix product state representations and the role of entanglement.
At its core DMRG is a variational optimization over the class of matrix product state (MPS) wavefunctions. An MPS expresses a many-body state as a product of site-local tensors, limiting the entanglement entropy according to an area-law for one-dimensional gapped systems. The method uses reduced density matrices to identify the most important states of a subsystem: truncating to the largest eigenvalues of the subsystem reduced density matrix yields an optimal low-rank approximation in the sense of the Schmidt decomposition. This truncation is conceptually linked to the renormalization group flow of relevant degrees of freedom and to measures such as the von Neumann entanglement entropy and Schmidt values. Theoretical analyses by Ian Affleck-related groups and Michael Levin clarified scaling and critical behavior, while connections to tensor network theory unified DMRG with higher-dimensional generalizations.
Practical DMRG proceeds by building a system and environment block, embedding sites, and performing iterative sweeps that optimize local tensors. Typical steps include: initializing a small superblock, diagonalizing the superblock Hamiltonian often via the Lanczos algorithm or Davidson algorithm, forming reduced density matrices for subsystems, truncating to a fixed bond dimension (number of kept states), and updating block representations. Two canonical variants are the infinite-system DMRG for growing chains and the finite-system DMRG with sweeping for converged energies. Observables are computed using effective operators within the truncated basis. Implementations frequently exploit symmetries such as U(1) or SU(2) to conserve quantum numbers and reduce computational cost; groups at Oak Ridge National Laboratory and software packages like ALPS project and ITensor supply optimized libraries.
DMRG has been extended to time-dependent problems (time-dependent DMRG, tDMRG) and real-time dynamics using schemes like time-evolving block decimation (TEBD) and matrix product operator (MPO) representations. Finite-temperature properties are obtained via purification techniques or the transfer-matrix DMRG and minimally entangled typical thermal states (METTS). Higher-dimensional and critical systems motivated tensor network generalizations such as projected entangled pair states (PEPS) and the multiscale entanglement renormalization ansatz (MERA) introduced by Guifre Vidal. Hybrid algorithms combine DMRG with density functional theory (DFT) in quantum chemistry, leading to the DMRG-CAS (complete active space) approaches developed by teams at CRAY Research and university chemistry departments.
DMRG has become the method of choice for one-dimensional correlated systems: detailed phase diagrams and correlation functions have been obtained for the Heisenberg chain, t-J model, Hubbard model, spin ladders, and disordered systems exhibiting many-body localization (MBL). In quantum chemistry, DMRG provides accurate descriptions of strongly correlated molecules and transition metal complexes where conventional coupled cluster methods fail. Condensed-matter studies using DMRG informed understanding of topological order in spin chains, edge states in Haldane phase, and entanglement spectra linked to conformal field theory predictions at criticality. Experimental relevance stems from cold-atom realizations in MIT and Max Planck Institute of Quantum Optics experiments that emulate lattice Hamiltonians solvable by DMRG.
Efficient DMRG implementations hinge on sparse linear algebra, optimized tensor contractions, and exploitation of symmetries. Computational cost scales roughly as O(m^3) to O(m^4) per step where m is the bond dimension, and memory scales as O(m^2). Parallelization strategies distribute blocks or quantum-number sectors; GPU acceleration has been explored by groups at NVIDIA and national labs. Benchmarks show excellent convergence for gapped one-dimensional systems, while critical or highly entangled states require much larger m. Public codes include ITensor, TeNPy, and components in the ALPS project.
DMRG's principal limitations are in higher dimensions and for systems with volume-law entanglement, where required bond dimensions grow exponentially. Extending DMRG to two-dimensional lattices motivates PEPS and tensor-network contraction challenges; approximate contraction schemes and Monte Carlo hybrids remain active research areas. Other frontiers include real-time dynamics at long times, finite-temperature scaling near critical points, incorporation with ab initio methods in quantum chemistry, and automated exploitation of non-Abelian symmetries. Ongoing work at institutions such as Perimeter Institute, Harvard University, and national laboratories continues to expand DMRG's reach and integrate insights from quantum information theory and machine learning.
Category:Numerical methods in quantum mechanics Category:Quantum many-body theory