| DMRG | |
|---|---|
| Name | Density Matrix Renormalization Group |
| Developer | Steven R. White |
| Introduced | 1992 |
| Field | Computational condensed matter physics |
| Related | Renormalization group, Tensor network |
DMRG
DMRG (Density Matrix Renormalization Group) is a numerical variational technique for obtaining ground states and low-energy properties of low-dimensional quantum many-body systems. Developed in the early 1990s, it revolutionized calculations for one-dimensional quantum spin chains and fermion models, and remains influential in studies of strongly correlated electrons, quantum information, and computational condensed matter physics.
DMRG was introduced by Steven R. White in 1992 to overcome limitations of traditional real-space renormalization group approaches for quantum lattice models. Early demonstrations focused on the Heisenberg model and the Hubbard model, showing dramatic improvements in accuracy for ground-state energies and correlation functions. The method emerged during a period when numerical approaches such as exact diagonalization and quantum Monte Carlo faced severe size or sign-problem constraints; DMRG provided a stable, controlled alternative for quasi-one-dimensional systems. Influential subsequent work connected DMRG to concepts from quantum information theory (notably entanglement entropy) and to tensor-based representations developed at institutions such as the Max Planck Institute for the Physics of Complex Systems and universities including University of California, Irvine and Massachusetts Institute of Technology.
At its core, DMRG is a variational algorithm that builds an efficient reduced representation of a many-body wavefunction by iterative truncation of the Hilbert space using the reduced density matrix of a subsystem. The algorithm splits a lattice into blocks and environments, diagonalizes the block density matrix, and retains eigenstates with largest eigenvalues to form a truncated basis. Key concepts include the use of reduced density matrix spectra, the Schmidt decomposition, and control of truncation error via retained eigenvalue weight. The original "infinite-system" and later "finite-system" DMRG sweeping protocols provide systematic improvement. Connections to matrix product states (MPS) render DMRG as a variational optimization within the MPS manifold; this link clarifies the success of DMRG for systems with limited entanglement and motivates performance measures based on entanglement entropy and the area law.
DMRG has been applied extensively to models central to condensed matter and quantum magnetism: the spin-1/2 Heisenberg chain, the Majumdar–Ghosh model, the Hubbard model in one dimension, the t-J model, and ladder systems such as two-leg spin ladders. It provides accurate results for ground-state energies, excitation gaps, correlation functions, and dynamical response when combined with correction-vector or time-dependent techniques. DMRG informs understanding of quantum phase transitions (e.g., between Luttinger liquid and gapped phases), topological order exemplified in the AKLT model and symmetry-protected phases, and quasi-one-dimensional realizations in materials studied at laboratories like Oak Ridge National Laboratory and Los Alamos National Laboratory. Beyond solid-state models, DMRG variants address problems in quantum chemistry (molecular electronic structure), implemented by groups at institutions such as Harvard University and ETH Zurich.
Originally presented as a form of real-space renormalization group adapted to quantum problems, DMRG differs from Wilsonian RG by retaining states based on density-matrix weight rather than energy truncation. The conceptual synthesis with quantum information theory and tensor network representations led to recognition of DMRG as an optimal algorithm within the matrix product state class. This places DMRG alongside related tensor methods such as projected entangled pair states (PEPS) for two dimensions and multiscale entanglement renormalization ansatz (MERA) for critical systems. The interplay between DMRG and tensor networks clarifies limitations: exponential growth of required bond dimension with two-dimensional area law violations and with critical entanglement, motivating hybrid strategies and approximations.
Practical DMRG implementations are available in many software packages and research codes developed at universities and companies, including ITensor (originally from Carnegie Mellon University and University of Wyoming collaborators), ALPS project, and quantum chemistry suites integrating DMRG modules. Performance considerations include choice of basis, exploitation of abelian and non-abelian symmetries (e.g., SU(2) symmetry), use of sparse linear algebra libraries, and parallelization strategies for large bond dimensions. Time-dependent variants such as time-dependent DMRG (tDMRG) and time-evolving block decimation (TEBD) require careful control of time-step error and bond growth; matrix product operator (MPO) representations are used for Hamiltonians and observables. Memory and CPU cost scale with the third or fourth power of the retained bond dimension in typical implementations, making algorithmic optimizations and symmetry exploitation critical for tractable simulations.
DMRG has spawned numerous extensions: DMRG-based dynamical methods (correction-vector, real-time evolution), finite-temperature DMRG via purification and minimally entangled typical thermal states (METTS), and applications to open quantum systems via Lindblad master equations. Advances connect DMRG to tensor network renormalization and to quantum simulation platforms; recent work explores cross-fertilization with quantum computing algorithms for quantum chemistry and materials, and scaling DMRG-like algorithms to two dimensions via PEPS and hybrid Monte Carlo–tensor approaches. Ongoing research addresses entanglement growth control, adaptive compression, and integration with experimental probes from facilities such as CERN-affiliated collaborations and national laboratories, ensuring DMRG remains a central, reliable tool for conservative, cumulative progress in theoretical and computational quantum physics.
Category:Computational physics Category:Condensed matter physics Category:Numerical linear algebra