Les Houches Lectures Notes on Tensor Networks

Abstract: Tensor networks provide a powerful new framework for classifying and simulating correlated and topological phases of quantum matter. Their central premise is that strongly correlated matter can only be understood by studying the underlying entanglement structure and its associated (generalised) symmetries. In essence, tensor networks provide a compressed, holographic description of the complicated vacuum fluctuations in strongly correlated systems, and as such they break down the infamous many-body exponential wall. These lecture notes provide a concise overview of the most important conceptual, computational and mathematical aspects of this theory.

• The paper presents a mathematically rigorous framework using tensor networks to capture entanglement and simulate quantum many-body phases.
• It details the use of matrix product states and MPO symmetries in algorithms like DMRG and VUMPS, emphasizing variational calculus and the geometry of quantum states.
• It further classifies topological and SPT orders through group cohomology and categorical symmetries, paving the way for efficient tensor network algorithms.

Authoritative Summary of "Les Houches Lectures Notes on Tensor Networks" (2512.24390)

Overview and Motivation

This lecture note series presents a thorough, mathematically rigorous exposition of tensor networks as the central framework for simulating and classifying correlated and topological phases of quantum matter. Beginning with a discussion of the exponential growth—more precisely, the double-exponential inaccessibility—of many-body Hilbert space, the authors argue that physically realizable states form a low-complexity manifold, with tensor networks capturing the entanglement structure responsible for this reduction. The lectures delineate both the variational manifold perspective and categorical structures underlying modern understanding of quantum phases.

Matrix Product States and Manifolds

A central conceptual development is the representation of physical quantum many-body ground states as points on the manifold of matrix product states (MPS). The authors systematically develop the geometry of the MPS manifold, establishing the utility of the time-dependent variational principle (TDVP) within this constrained subspace and formalizing its symplectic structure. They emphasize that DMRG and its descendants are specific numerical instantiations of energy minimization on the MPS manifold, highlighting the formal connection via Poisson brackets and variational calculus.

Monogamy of entanglement is placed at the core of why mean-field approximations fail in 1D, and why area laws emerge for gapped systems. The manifold approach is compared concretely with other variational submanifolds (e.g., Slater determinants for fermions, coherent states for bosons), clarifying the superior compressive power of MPS for capturing essential strong correlations in low dimensions.

Tensor Networks, Area Laws, and Algorithms

The lectures give a precise technical treatment of area laws, explicitly bounding the entanglement entropy of an MPS by $\log \chi$ (bond dimension) and referencing the Hastings area law for gapped chains. The construction and graphical notation for MPS and their tangent vectors is detailed, as well as the related class of matrix product operators (MPOs).

Algorithmically, both finite (DMRG) and infinite system (\texttt{VUMPS}) ground state search protocols are presented through the unifying variational principle lens. Normal forms and gauge freedoms of MPS are exploited to improve computational scaling and to define (quasi-)particle excitation ansätze within the tangent space (building on Feynman-Bijl structure). The lecture carefully attaches the decay of correlations to the spectral gap in the MPS transfer operator, and not merely to locality.

Critical systems, where area laws are logarithmically corrected, are dealt with via finite correlation length scaling, showing deep connections to CFT exponents and central charges extracted numerically from finite-$\chi$ MPS data. This presentation is rigorous and connects with current best practices in scaling analysis for tensor-network based algorithms.

Quantum Phases, Symmetry, and SPT Orders

One of the strongest contributions of the notes is the categorical approach to quantum phases. The Landau paradigm is generalized, emphasizing that in 1D, gapped phases with global symmetry $G$ are classified by (i) the unbroken subgroup $H \leq G$, and (ii) a projective representation (cocycle class in $H^2(H, U(1))$) dictating SPT order—these statements are derived explicitly from the fundamental theorem of MPS, which is proved constructively.

The connection between entanglement (Schmidt) spectrum degeneracies, virtual representation structure, and edge mode robustness in SPT phases is analyzed in depth, making strong predictions about physical observables (minimally degenerate multiplets determined by cocycle classes). All symmetry classes—including anti-unitary actions (time-reversal, reflection)—are treated with full technical generality using group cohomology with twisted action, naturally leading to new sets of invariants ($H^2(G, U(1)^{\Theta})$).

Fermionic tensor networks are also treated systematically, with detailed attention to the supervector space structures, grading, and the subtleties of even/odd parity projectors. The notes explicate the construction of fermionic MPS (fMPS), the role of twisted boundary conditions, and the nontrivial SPT classification in interacting fermionic systems (recovering the Fidkowski-Kitaev $Z_8$).

Higher Dimensions, MPO Algebras, and Topological Order

The PEPS formalism as a higher-dimensional (2D, 3D) generalization of MPS is treated from a categorical algebra perspective. The notes conjecture a (now widely accepted and numerically substantiated) "fundamental theorem of PEPS": two PEPS represent the same state for all system sizes if and only if they are intertwined by an MPO symmetry, extending the MPS gauge transformation.

A key result is the construction and classification of topologically ordered phases (e.g., quantum double/toric code, string-net models) via PEPS with nontrivial MPO symmetries. Here, virtual symmetry sectors, fusion categories, and tube algebra idempotents are shown to label ground state degeneracies and anyon excitations, connecting tensor networks directly to the data of modular tensor categories and the Drinfeld center.

The treatment of 2D SPT phases via 3-cocycles $H^3(G, U(1))$ is also explicit, and a graphical calculus is developed for symmetry fusion tensors. The links between bulk and boundary—edge Hamiltonians with inherited (possibly anomalous) MPO symmetries—are analyzed categorically, and the obstruction to gapped symmetric boundaries is identified as a nontrivial $H^3(G, U(1))$ anomaly.

Generalized Symmetries, Dualities, and Categorical Structure

The lecture culminates in the modern theory of generalized (categorical) symmetries, describing MPO symmetries as representations of fusion categories (beyond group actions). Module categories classify phases, with Morita equivalence relating dual tensor categories, and the pentagon equation on $F$-symbols encoding all associativity constraints of fusion.

Applications include:

• The explicit construction of string-net ground states,
• The "strange correlator" as a general CFT/TQFT correspondence on the lattice,
• A rigorous algebraic theory of dualities (Kramers-Wannier, Kennedy-Tasaki), each realized as specific MPO intertwiners related to module categories over the symmetry fusion category.

Operators and Hamiltonians invariant under MPO symmetry are classified via generalized Clebsch-Gordan (fusion) tensors. The structure constants of these bond algebras depend only on the categorical data, and spectra are independent of the module category, leading to a precise web of dualities (with potentially distinct Hilbert space constraints).

The generalization of the Landau paradigm is formulated precisely: gapped phases with categorical symmetry $\mathcal{C}$ are in bijection with module categories over $\mathcal{C}$. The existence of a completely symmetric phase is determined by the triviality of the anomaly, and entanglement structure is shown to be minimized in the fully symmetry-broken dual.

Implications and Future Directions

This compendium makes several important theoretical predictions and consolidates the categorical language as the future foundation for work in quantum phases and quantum information. Practically, these results inform the construction of optimal tensor network algorithms (e.g., for DMRG in symmetry-broken duals), and provide routes to engineer topological models with prescribed anyon content.

Conceptually, the lectures elucidate open problems in:

• Generalizing PEPS fundamental theorems,
• Constructing tensor network representations for arbitrary fusion 2-categories/higher-form symmetries,
• Realizing topological/categorical dualities directly in lattice Hamiltonians (gauging, symmetry-enriched constructions),
• Extending the formalism and numerics to continuous symmetries and higher dimensions.

The categorical duality and symmetry-breaking framework developed forms the mathematical infrastructure for both new theoretical models of quantum matter and the design of numerically efficient tensor network algorithms.

Conclusion

"Les Houches Lectures Notes on Tensor Networks" (2512.24390) offers an exhaustive, meticulous, and deeply technical synthesis of tensor networks and their foundational role in quantum many-body theory and statistical physics. It advances the categorical approach to symmetry, duality, and phase structure, linking computational advances directly to deep mathematical structure. The lecture notes serve both as a technical manual for state-of-the-art simulation techniques and as a blueprint for research at the intersection of mathematical physics, condensed matter, and quantum information science.