Variational Ground State Optimization with Matrix Product States

19 concepts19 papers1 proposal

The central challenge in quantum many-body physics is the curse of dimensionality. For a system of $N$ interacting particles, each capable of occupying $d$ local states, the size of the total Hilbert space scales as $d^N$. For even a modest number of particles, storing the full wave function in memory becomes impossible, rendering exact diagonalization useless.

Yet, nature routinely finds the ground states of these complex systems. This suggests that physical ground states occupy a tiny, highly structured corner of the full Hilbert space. The Density Matrix Renormalization Group (DMRG) and its modern formulation via Matrix Product States (MPS) exploit this structure. By leveraging the physical principle of the entanglement area law, this framework compresses the exponential complexity of the Hilbert space into a polynomially scalable variational optimization problem.

The Area Law and the Matrix Product Ansatz

Why can we compress a quantum wave function? The answer lies in how quantum entanglement is distributed in physical ground states.

If you choose a random state in the Hilbert space and divide the system into two subsystems, the entanglement entropy between them scales with the volume of the subsystems. However, for ground states of local, gapped Hamiltonians, the entanglement entropy scales only with the area of the boundary separating them. In one dimension (1D), the boundary between two halves of a chain is just a single point (an area of a constant size). Consequently, the entanglement entropy remains bounded by a constant, regardless of how long the chain grows.

An MPS builds this area law directly into the mathematical structure of the wave function. Instead of describing the state using a single tensor with $N$ indices, $C_{i_1 i_2 \dots i_N}$, we factorize it into a chain of matrix multiplications:

$$|\psi\rangle = \sum_{i_1, \dots, i_N} A^{i_1} A^{i_2} \dots A^{i_N} |i_1, i_2, \dots, i_N\rangle$$

Here, each $A^{i_k}$ is a matrix of size $\chi \times \chi$ for each local physical state $i_k$. The parameter $\chi$ is known as the bond dimension. The bond dimension acts as a strict throttle on entanglement: the maximum entanglement entropy that can flow through a bond is $\ln(\chi)$. Because the area law guarantees that the required entanglement is bounded in 1D, we can keep $\chi$ fixed and scale the system size $N$ linearly without losing accuracy.

The DMRG Optimization Mechanism

Knowing that an MPS can represent the ground state is only half the battle; we also need an efficient way to find the specific matrices that minimize the system's energy. DMRG achieves this through a highly efficient local sweeping procedure.

Instead of optimizing all matrices simultaneously—which would reintroduce non-linear complexities—DMRG optimizes one or two adjacent sites at a time while keeping the rest of the chain fixed. The algorithm projects the global Hamiltonian into an effective environment basis formed by the fixed sites. This reduces the daunting global problem to a standard, low-dimensional local eigenvalue problem that can be solved using iterative solvers like the Lanczos or Davidson algorithms.

Once the local state is updated, the algorithm moves to the next site. To maintain the MPS structure and control the growth of the bond dimension, a Singular Value Decomposition (SVD) is performed on the updated local tensor:

$$M = U \Sigma V^\dagger$$

The diagonal matrix $\Sigma$ contains the singular values, which correspond directly to the square roots of the eigenvalues of the reduced density matrix. These values quantify how much each quantum state contributes to the entanglement between the left and right halves of the system.

To prevent the bond dimension from growing indefinitely, we truncate the matrices by keeping only the $\chi$ largest singular values. Because the singular values typically decay exponentially for area-law states, discarding the smallest values introduces a negligible truncation error. The algorithm then "sweeps" back and forth across the lattice, iteratively minimizing the energy until the wave function converges to the true ground state.

For infinite systems where boundary effects must be entirely eliminated, variations of this approach are used. For instance, Rui-Zhen Huang et al. utilized a variational MPS algorithm within the tangent space to determine the ground state of infinite spin chains with high precision, capturing emergent symmetries at quantum critical points.

Extending to Quasi-2D Geometries

While the area law makes MPS perfectly suited for 1D chains, many of the most compelling mysteries in condensed matter physics—such as high-temperature superconductivity and fractional quantum Hall states—occur in 2D. In 2D, the boundary area scales with the width of the system ($L_y$), meaning the entanglement entropy grows linearly with $L_y$, and the required bond dimension $\chi$ scales exponentially as $e^{L_y}$.

Despite this exponential barrier, DMRG remains remarkably effective for quasi-2D geometries by mapping a 2D lattice onto a 1D snake-like path. By keeping the width $L_y$ finite and scaling the length $L_x$ to infinity, researchers can rigorously probe 2D physics.

Quasi-2D Cylinder Mapping to a 1D MPS Path:
   ┌───>───┐   ┌───>───┐
   │ 1   2 │   │ 5   6 │   ...
   └───┐   └───┘   ┐   │
       ▼           ▼   ▼
   ┌───┴───┐   ┌───┴───┐
   │ 4   3 │   │ 8   7 │   ...
   └───────┘   └───────┘
  (Periodic boundary conditions along Y)

This geometric mapping takes several forms depending on the physical system:

  • Multi-leg Ladders: Hao-Kai Zhang et al. applied DMRG to multi-leg $t$-$J$ ladders to isolate the non-local phase string effects responsible for spatial oscillations in superconductivity. Similarly, Steven A. Kivelson et al. simulated a multi-orbital diagonal ladder model to resolve competing magnetic and superconducting phases in iron-pnictide superconductors.
  • Cylindrical Geometries: By wrapping one dimension into a tube with periodic boundary conditions, researchers can eliminate edge effects. Yiqing Zhou et al. used large-scale cylindrical DMRG to show how long-range Coulomb interactions stabilize chiral spin liquids in transition metal dichalcogenide (TMD) moiré systems. In subsequent work, Yiqing Zhou et al. utilized this setup to pinpoint the first-order nature of the quantum melting transition of Wigner crystals into Fermi liquids.
  • Infinite Cylinders (iDMRG): To study topological phases without longitudinal boundary interference, the infinite DMRG framework can be extended to cylinders. For example, Julian Boesl et al. optimized an infinite MPS on a cylinder to characterize fractional topological phases and particle-hole symmetry in interacting lattice bosons.

Unveiling Exotic Quantum Phenomena

Because DMRG provides an unbiased, numerically exact window into strongly correlated systems where standard mean-field theories fail, it has become the premier tool for discovering and validating exotic phases of matter.

Hidden Non-Local Orders

Some quantum phases do not possess a conventional local order parameter. The Haldane Bose Insulator is a prime example, characterized by a hidden string order where fluctuations are non-local. Emanuele Dalla Torre et al. used the precision of DMRG to calculate these non-local correlation functions and map the phase boundaries of 1D ultracold atomic gases. Further exploration by Emanuele G. Dalla Torre et al. demonstrated how local perturbations and inter-chain couplings can systematically dismantle this hidden string order.

Transport and Criticality

DMRG also excels at calculating ground-state properties that dictate transport behavior. By simulating large chains with hundreds of sites, Xiaoqun Wang et al. demonstrated that spin transport in Heisenberg antiferromagnetic chains depends entirely on the spin quantum number, showing ballistic flow for $S=1/2$ but exponential decay for $S=1$. When analytical mappings to non-interacting fermions are impossible, such as in models featuring Dzyaloshinskii-Moriya interactions, DMRG provides the necessary numerical recourse to map out sustainable entanglement dynamics, as shown by Titas Chanda et al..

Superconductivity and Correlated Insulators

In the search for high-temperature superconductivity, DMRG serves as a definitive arbiter between competing ground states. Jordan Venderley provided the first robust evidence of an elusive pair-density wave (PDW) state in spin-valley locked systems, demonstrating that the PDW can be energetically favored over uniform superconductivity.

The algorithm's ability to handle strong electron-electron interactions also allowed Li Liu et al. to map the phase diagram of the infinite-$U$ Hubbard model on square lattices. Furthermore, it enabled Jiaxin Zhang et al. to identify an emergent $\mathbb{Z}_2$ Berry phase that drives interlayer Cooper pairing in nickelate superconductors.

Beyond Condensed Matter: Quantum Chemistry

The utility of DMRG extends beyond idealized lattices into quantum chemistry. In complex molecules, electrons occupy orbitals with highly non-local interactions, yet the underlying entanglement structure remains compressible. Matthieu Mottet et al. demonstrated that by computing single-orbital entropy and mutual information via DMRG, quantum entanglement can be used to provide a rigorous definition of chemical bonds and accurately predict the breaking mechanisms of covalent bonds where standard mean-field approaches fail.

Whether tracking the melting of Wigner crystals in arrays of 1D tubes (Michael Knap et al.) or evaluating the destabilization of an Ising critical point coupled to a gapless boson (Ori Alberton et al.), the core mechanism remains identical. By compressing the wave function according to its physical entanglement structure, DMRG transforms an intractable exponential problem into a controlled, linear sequence of local matrix optimizations.

Key Foundational Work

Hidden Order in 1D Bose Insulators (2006) — contributed 1 concept elements to this direction.
Spin transport properties in Heisenberg antiferromagnetic spin chains: Spin current induced by twisted boundary magnetic fields (2006) — contributed 1 concept elements to this direction.
Rise and fall of hidden string order of lattice bosons (2008) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_density_matrix_renormalization_group_2b51ea2f,
  title = {density matrix renormalization group},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/density-matrix-renormalization-group/rdr_2b51ea2fe8144af28c91cf1620991197},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-22},
}

This exact version (version 1, frozen text):

@misc{lacuna_density_matrix_renormalization_group_2b51ea2f_v1,
  title = {density matrix renormalization group},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/density-matrix-renormalization-group/rdr_2b51ea2fe8144af28c91cf1620991197/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-22},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake