| DMRG | |
|---|---|
| Name | Density matrix renormalization group |
| Acronym | DMRG |
| Developer | Steven R. White |
| Introduced | 1992 |
| Field | Condensed matter physics |
| Related | Matrix product state, Renormalization group |
DMRG
DMRG is a numerical variational technique for finding low-energy states of strongly correlated quantum many-body systems. It is particularly effective for one-dimensional lattice models and is widely used in Condensed matter physics, Quantum chemistry, and the study of Quantum information. DMRG matters because it provides highly accurate approximations to ground states and excitations where exact diagonalization or perturbation theory fail.
Originally developed to overcome limitations of real-space Renormalization group approaches for lattice systems, DMRG constructs optimized truncated bases using reduced density matrices to preserve relevant quantum correlations. It has become a standard tool for studying models such as the Heisenberg model, the Hubbard model, and the t-J model, enabling quantitative predictions for spin chains, fermionic chains, and interacting bosons. DMRG's precision in calculating energies, correlation functions, and entanglement measures established connections to Quantum information theory and the theory of Matrix product state ansätze.
DMRG was introduced by Steven R. White in 1992 to address failures of traditional real-space renormalization for low-dimensional quantum lattices. Early benchmarking compared DMRG results to exact solutions like the Bethe ansatz for the Heisenberg chain and to numerical exact diagonalization on small clusters. Foundational work linked DMRG to the Schmidt decomposition and reduced density matrix eigenvalue spectra; key conceptual advances were provided by researchers including Ulrich Schollwöck, Ian Affleck, and Frank Verstraete, who clarified the connection to Matrix product states and Tensor networks.
The core DMRG algorithm iteratively grows a system block and an environment block, performing a variational optimization by diagonalizing the superblock Hamiltonian and truncating to the leading eigenvectors of the reduced density matrix. Practical implementations rely on sparse eigensolvers such as the Lanczos algorithm and Davidson algorithm and exploit symmetries (e.g., U(1), SU(2)) to reduce computational cost. Modern codes are available in libraries and packages maintained by groups at institutions like Max Planck Institute for Physics of Complex Systems, University of California, Irvine, and projects such as ITensor and ALPS (software). Numerical implementation details include choice of boundary conditions, sweep schedules (infinite and finite DMRG), truncation dimension (bond dimension), and measures of convergence.
DMRG can be understood as a variational method over the class of Matrix product state (MPS) wavefunctions; the density matrix truncation corresponds to truncating the MPS bond dimension. This interpretation unifies DMRG with the language of Tensor network states and enables systematic generalizations to algorithms such as Time-evolving block decimation (TEBD) and Projected entangled pair states (PEPS). The MPS framework explains why DMRG excels for gapped one-dimensional systems obeying an area law for entanglement, and why it encounters difficulties for high-entanglement states such as critical systems described by Conformal field theory.
DMRG has been applied extensively to compute ground states, excitation spectra, and dynamical correlation functions of models including the Heisenberg model, the Hubbard model, the Kondo model, and the Holstein model. It is used to study quantum phase transitions, spin ladders, and impurity problems relevant to Kondo physics and quantum transport. Extensions and cross-disciplinary applications include quantum chemistry calculations for molecular electronic structure, studies of cold atoms in optical lattices (experiments at institutions like CERN and MIT), and modeling of quasi-one-dimensional materials such as SrCuO2 and organic conductors.
DMRG's efficiency is tied to the scaling of entanglement entropy: for one-dimensional gapped systems that satisfy an area law, a modest bond dimension suffices for accurate results. For critical systems with logarithmic entanglement growth (described by Conformal field theory), bond dimensions must increase to maintain accuracy. In higher dimensions or for highly excited states, the entanglement growth renders DMRG computationally demanding; alternative tensor network ansätze like PEPS or Monte Carlo methods such as Quantum Monte Carlo are often preferable. Finite-size effects, choice of truncation cutoff, and preservation of symmetries are practical limitations; diagnostics include discarded weight and spectrum of reduced density matrices.
Time-dependent formulations such as time-dependent DMRG (t-DMRG) and TEBD enable simulation of unitary dynamics, quenches, and transport, often using Suzuki–Trotter decompositions or Krylov subspace methods. Finite-temperature extensions use purification, minimally entangled typical thermal states (METTS), and ancilla techniques to compute thermal expectation values. Efforts to extend DMRG to two dimensions include snake-like mappings of 2D lattices to 1D chains, development of PEPS, and hybrid approaches combining DMRG with Dynamical mean field theory (DMFT) or Density functional theory (DFT). Recent algorithmic improvements incorporate matrix product operator (MPO) representations, adaptive time-stepping, exploitation of non-Abelian symmetries, and GPU-accelerated implementations; these advances are driven by research groups at University of Bonn, University of Innsbruck, ETH Zurich, and national laboratories such as Lawrence Berkeley National Laboratory.
Category:Computational physics Category:Condensed matter physics