Infinite Matrix Product State Simulations

• Infinite matrix product states (iMPS) are defined by a repeating tensor network structure that enables the simulation of quantum many-body systems directly in the infinite limit.
• They utilize canonical forms and fixed-point transfer matrices to compute observables and entanglement measures, eliminating finite-size artifacts.
• Advanced extensions with infinite MPOs and Grassmann techniques allow precise studies of real-time dynamics, phase transitions, and impurity problems while managing computational complexity.

An infinite matrix product state (iMPS) is an ansatz and computational framework for simulating quantum many-body systems directly in the thermodynamic (infinite-system) limit using tensor networks with a repeating unit-cell structure. Infinite MPS simulations have enabled precision studies of ground-state properties, entanglement, correlation functions, dynamics, phase transitions, and quantum transport in both lattice models and impurity problems, even in regimes where standard exact diagonalization or finite-size MPS methods are intractable. The methodology encompasses a spectrum of implementations, including canonical forms on separable infinite-dimensional Hilbert spaces, approaches tailored to non-equilibrium and steady-state dynamics using infinite Grassmann MPS, and advanced boundary-free strategies for simulating large or truly infinite systems without finite-size or edge artifacts.

1. Fundamental iMPS Structures and Canonical Forms

The core of an infinite MPS simulation is a tensor network state

$$|\Psi\rangle = \cdots A^{\sigma_{n-1}} A^{\sigma_n} A^{\sigma_{n+1}} \cdots|\ldots\,\sigma_{n-1} \sigma_n \sigma_{n+1} \ldots\rangle,$$

with a local tensor $A^\sigma_{ij}$ and fixed (possibly site-dependent for larger unit cell) bond dimension, repeated endlessly along the 1D lattice. For translation-invariant states, canonical forms can be imposed: left-canonical ($\sum_\sigma (A^\sigma)^\dagger A^\sigma = I$), right-canonical ($\sum_\sigma A^\sigma (A^\sigma)^\dagger = I$), or mixed (with a central diagonal Schmidt tensor). In infinite-dimensional Hilbert spaces, every state in the tensor product of separable Hilbert spaces admits such a (possibly infinite-bond) decomposition using singular value decompositions of compact operators; canonical left- and right-normalized forms as well as mixed forms are available with orthonormality relationships across bipartitions holding analogously to the finite-dimensional case (Heikkinen, 18 Feb 2025).

The construction is not limited by the local Hilbert space dimension: idMPS methods rigorously translate the MPS architecture to settings such as quantum harmonic chains with infinite-dimensional local basis, with convergence and error controlled by truncations in the singular spectrum.

2. Algorithmic Methods and Stationarity in the Infinite Limit

The infinite limit is handled via explicit fixed-point equations for the so-called transfer matrix: $T(X) = \sum_\sigma A^\sigma X (A^\sigma)^\dagger,$ with the iMPS state characterized by dominant left and right eigenoperators ($T(R)=R$, $T^*(L)=L$), setting the canonical gauge for normalization and expectation values. All observables, including one- and two-point correlation functions, are computed as contractions with these boundary fixed points, yielding thermodynamic-limit values free of edge or finite-size effects (Phien et al., 2012, Michel et al., 2010).

Time evolution and real- or imaginary-time dynamics leverage the time-evolving block decimation (iTEBD) framework or variational approaches (iDMRG, VUMPS), exploiting translation invariance to update only a small repeating subset of tensors. For steady-state computations—e.g., in quantum impurity problems—stationarity is enforced directly by requiring that application of the infinite MPO representation of the evolution operator (detailed below) leaves the iMPS fixed. This leads to fixed-point superoperator equations whose dominant eigenoperators govern both boundary conditions and observable extraction (Guo et al., 2024).

3. Infinite MPOs, Grassmann Path Integrals, and Advanced Extensions

Complex infinite MPS algorithms extend to systems with non-local structures and fermionic degrees of freedom using powerful MPO (matrix product operator) and infinite Grassmann-MPS (G(i)MPS) constructions:

• Grassmann Infinite MPOs (iGTEMPO/iGMPS): Time-evolution for quantum impurity models, including non-equilibrium steady states and real-time Green's functions, can be computed using infinite Grassmann MPS representations of the augmented density tensor. The steady-state iGMPS is built as a single repeating unit-cell carrying all Grassmann variables for the forward/backward Keldysh contours, updated by action of an infinite Grassmann-MPO encoding the Feynman–Vernon influence functional (Guo et al., 2024, Sun et al., 2024). Structure and stationarity are maintained by contraction and SVD-based compression, with convergence to the fixed-point guaranteed by the transfer superoperator.
• Prony and Exponential Sum Decompositions: Key to computational tractability in these contexts is the decomposition of non-local kernels (e.g., hybridization functions) into truncated sums of exponentials, allowing for efficient MPO representations with moderate bond dimension (Guo et al., 2024, Guo et al., 2024).
• Window GMPS and Boundary Eigenproblems: For general time-dependent impurity problems, the path-integral is divided into semi-infinite "past," finite "window" (where the drive acts), and semi-infinite "future" regions, each described by infinite GMPS segments connected via boundary tensors. This allows the algorithmic cost to remain independent of the simulated time-window, as only the finite central region requires nonuniform treatment (Sun et al., 2024).
• Embracing Divergences: Infinite MPS methods can encode not only finite expectation values but also physically meaningful divergences (linear, polynomial, or exponential in system size), extracting exact scaling functions, their leading divergent terms, and finite corrections (Crosswhite, 2013).

4. Boundary-Condition Strategies: Infinite Boundary and Environment-MPO Embeddings

Several frameworks have been developed for accurately simulating the bulk properties of infinite systems without finite-size artifacts:

• Infinite Boundary Conditions (IBC): IBC techniques embed a finite window of interest between two boundary tensors encoding semi-infinite environments, matched to the fixed points of the bulk iMPS transfer matrix. This allows exact inclusion of infinite environments for local dynamics (e.g., time-evolution of local perturbations, computation of spectral functions) while using standard finite-MPS algorithms within the window (Phien et al., 2012). This formalism overcomes limitations of brute-force finite-size calculations, eliminating artificial reflection and Friedel oscillations.
• Environment-MPO Recursive Embedding: A more general boundary-free approach constructs effective MPOs for the Hamiltonians of semi-infinite regions directly from the finite MPS ground state via successive contraction and truncation. These environment MPOs are then iteratively attached to new finite windows, driving the state toward bulk thermodynamic properties with negligible finite-size error. This method is agnostic to translation-invariance of the local Hamiltonian and suitable for inhomogeneous or impurity problems, and supports arbitrarily long real-time dynamics without back-reflection (Shimozono et al., 8 Dec 2025).

5. Physical Applications and Performance Benchmarks

Infinite MPS simulations have enabled high-precision studies across a variety of contexts:

• Ground States and Critical Scaling: For 1D gapped Hamiltonians, iMPS with controlled bond dimension approximation can uniformly approximate all local observables on $\ell$ sites to accuracy $\epsilon$ with bond dimension scaling as $D\sim(\ell-1)/\epsilon$, underpinned by area laws and rigorous decay of Schmidt coefficients (Schuch et al., 2017). This underlies both rigorous error control and practical algorithmic selection of bond dimension in DMRG, iDMRG, and iTEBD codes.
• Entanglement and Phase Transitions: Entropy scaling, cumulants (Binder ratios), and finite-entanglement scaling analysis can be performed directly in the iMPS limit, allowing extraction of central charge, critical exponents, and phase boundaries without finite-size extrapolation (Pillay et al., 2019). For systems with long-range interactions, ground-state entropies and correlation structure are quantitatively resolved (e.g., logarithmic scaling in spin chains with power-law couplings) (Li et al., 2015).
• Real-Time and Steady-State Quantum Transport: In steady-state quantum impurity models, iGTEMPO obtains real-time Green's functions, noise-free steady-state currents, and avoids both sign problems and bath discretization error, with a cost independent of measurement time and number of baths (Guo et al., 2024). Zero-temperature imaginary-time and general non-equilibrium real-time problems are also efficiently tractable with the infinite MPS path-integral methods (Guo et al., 2024, Sun et al., 2024).
• Gauge Theories and Symmetry-Constrained Systems: iMPS with symmetric tensor and link-enhanced MPO structures allow for direct simulation of infinite-lattice gauge theories, enforcing Gauss's law and accommodating both Abelian and non-Abelian settings (Dempsey et al., 22 Aug 2025).
• Quantum Computing Implementations: Infinite MPS optimization and time evolution algorithms (e.g., via TDVP) can be mapped to finite-depth circuit ansätze on small quantum processors, enabling parallel quantum simulation of systems much larger than the hardware size, with scaling set by bond dimension and entanglement rather than physical size (Barratt et al., 2020).

6. Computational Complexity and Scalability

All infinite MPS simulations share the feature that computational cost is asymptotically dominated by the bond dimension $\chi$ (or $D$), with basic contraction and update steps scaling as $O(\chi^3)$ for standard iMPS, or $O(w\,\chi^3)$ for observables over a window of width $w$ (Guo et al., 2024). The number of physical baths or system size does not directly impact cost once the transfer matrix's fixed point is reached, and advanced Prony/exponential-sum representations of nonlocal kernels limit the growth of MPO bond dimension. For Grassmann path-integral-based approaches, operator anti-commutation is handled via parity-graded tensor blocks, ensuring stability and efficiency of all contractions.

7. Challenges, Limitations, and Prospects

While infinite MPS methods deliver unparalleled access to bulk quantum physics, several technical challenges remain:

• Limitations in Critical and Highly-Entangled Regimes: While gapped area-law systems require only modest bond dimensions for accurate iMPS representations, critical and especially long-range entangled states need larger $\chi$, with analogously higher computational cost.
• Multi-Orbital and High-Complexity Systems: The cost of multi-orbital impurity Green's function evaluation grows exponentially with the number of orbitals ($O(\chi^{2M})$), limiting practical studies to small $M$ in existing implementations (Guo et al., 2024).
• Fermionic and Infinite-Dimensional Local Spaces: The infinite-dimensional canonical theory provides a foundation for continuous-variable and field-theoretic systems, but practical algorithms require implementing cutoffs in physical and auxiliary spaces and reusing robust algebraic properties of compact operator decomposition (Heikkinen, 18 Feb 2025).
• Boundary Embedding and Error Control: Newly developed approaches such as environment-MPO embedding or recursion require additional DMRG solves and careful truncation to environment dimension, and their scaling with increasing complexity is under continuous investigation (Shimozono et al., 8 Dec 2025).
• Beyond One Dimension: Extensions to two dimensions (e.g., PEPS, boundary-MPO embedding in 2D cylinders) are conceptually straightforward but face steep computational scaling challenges.

Despite these challenges, the landscape of infinite MPS simulation continues to rapidly expand, integrating innovations from operator theory, tensor optimization, quantum circuit realizations, and boundary-free recursive algorithms for truly large-scale quantum many-body physics (Guo et al., 2024, Shimozono et al., 8 Dec 2025, Barratt et al., 2020, Heikkinen, 18 Feb 2025).