Tensor Networks: Computational Guide

A tensor-network calculation is persuasive only when the chosen network, contraction method, optimization procedure, and physical extrapolation are all controlled. A compact state representation is the beginning of a computation, not its conclusion.

The evidence chain is

$$\begin{gathered} \text{physical target} \\ \downarrow \\ \text{network and ordering} \\ \downarrow \\ \text{controlled contractions} \\ \downarrow \\ \text{controlled optimization} \\ \downarrow \\ \text{convergence study} \\ \downarrow \\ \text{qualified claim}. \end{gathered}$$

This guide helps choose and audit that chain. It deliberately does not redevelop the tensor definitions and entanglement derivations that already have canonical homes.

Purpose and canonical scope

This route is an intentional computational bridge.

Tensor Networks Preview is the canonical conceptual article. It owns the graph and index dictionary, virtual gauge freedom, cut-capacity bounds, MPS–PEPS–MERA comparison, contraction geometry, and the distinction among representation, contraction, and optimization.

Matrix Product States Preview owns one-dimensional MPS forms, Schmidt bonds, canonical gauges, transfer operators, injectivity, finite-entanglement effects, and the conceptual MPS–DMRG relation.

Area Laws owns theorem status and the limits of the area-law argument. Entanglement Spectrum owns Schmidt tails and local truncation data.

This page adds only what a reader entering from computational many-body physics needs:

• a routing map from physical targets to network families;
• the four gates a finite-bond calculation must pass;
• convergence designs for chains, cylinders, PEPS, critical states, dynamics, and thermal states;
• an error ledger separating representation, contraction, optimization, and extrapolation;
• reporting standards and stopping criteria;
• compact exercises in auditing numerical claims.

DMRG Preview gives finite-system MPS optimization its own algorithmic treatment, including effective Hamiltonians, one-site and two-site sweeps, truncation, and convergence. Tensor-Network Simulation owns circuit-to-network translation, MPS circuit execution, spacetime contraction paths, slicing, sampling, and circuit-simulator validation. General production tensor libraries, data structures, and benchmark implementations remain the responsibility of the future Computational QM volume.

Canonical reading map

Use the shortest route that supplies the missing concept.

Need	Canonical destination
What a tensor-network graph means	Tensor Networks Preview: From One Tensor to a Network
Why virtual bonds constrain entanglement	The General Cut Bound
MPS, PEPS, tree, and MERA comparison	A Family Map
MPS canonical form and contractions	Matrix Product States Preview
Circuit contraction, slicing, and sampling	Tensor-Network Simulation
Schmidt tails and discarded weight	Entanglement Spectrum
Area-law theorem and converse cautions	Area Laws
Critical finite-entanglement scaling	Entanglement and Criticality
Projected tensor-manifold dynamics	Time-Dependent Variational Principle
General computational validation	Computational Many-Body Overview

Declare the computational target

Before choosing a network, write the task as

$$\mathcal P = \left( H, \mathcal G, \mathcal B, \mathcal S, \rho, O, \mathcal L, \epsilon_{\mathrm{target}} \right).$$

Here:

• $H$ is the Hamiltonian or evolution generator;
• $\mathcal G$ is the physical interaction geometry and site ordering;
• $\mathcal B$ is the boundary condition;
• $\mathcal S$ is the symmetry and topological sector;
• $\rho$ is the target state or ensemble;
• $O$ is the observable set;
• $\mathcal L$ is the physical limit, such as $L\to\infty$ or $t\to t_{\max}$;
• $\epsilon_{\mathrm{target}}$ is the desired accuracy under a declared metric.

The best network for a one-dimensional ground-state energy need not be best for a two-dimensional long-range correlator, a thermal density operator, or real-time evolution of the same Hamiltonian.

Four computational gates

A tensor-network result must pass four logically distinct gates.

Representability

Does the selected family contain a sufficiently accurate state at accessible bond dimension?

For a family $\mathcal M_D$, the best variational energy is

$$E_D^* := \inf_{\lvert\phi\rangle\in\mathcal M_D} \frac{ \langle\phi|H|\phi\rangle }{ \langle\phi|\phi\rangle }.$$

The existence of $E_D^*$ says nothing yet about whether it can be evaluated or found.

Evaluability

Can norms, energies, gradients, reduced states, and target observables be contracted with controlled error?

Chains permit exact sequential contraction at polynomial cost in the MPS bond dimension. Looped networks such as PEPS generally require an approximate environment or a cost that grows rapidly with width. The environment cutoff is therefore a convergence variable distinct from the state bond dimension.

Searchability

Can the optimization reach a good point in the chosen family?

Let the returned energy be $E_D^{\mathrm{found}}$. Then

$$\delta E_{\mathrm{opt}} := E_D^{\mathrm{found}} - E_D^* \ge 0.$$

Small changes during the final sweep or iteration show local stationarity, not that $\delta E_{\mathrm{opt}}$ is small. Restarts, symmetry sectors, unit cells, noise schedules, update type, and initialization can expose optimization dependence.

Inferential control

Does the finite calculation support the claimed physical limit and observable?

Even an exact optimum inside $\mathcal M_D$ can be a poor approximation to the intended thermodynamic, continuum, long-time, or critical result. Bond dimension must be varied alongside size, width, time step, boundary condition, and every other active cutoff.

Family selection

Network names describe state families and contraction geometries. Optimization algorithms are separate choices.

Family or workflow	Natural first target	Main strength	Decisive controls
finite MPS with DMRG	low-energy states of finite 1D chains	exact contractions and strong Schmidt compression	$L$, $\chi$, sweeps, restarts, sector, boundary
uniform or infinite MPS	translation-invariant 1D bulk states	direct thermodynamic representation	$\chi$, unit cell, transfer fixed point, initialization
MPS time evolution	short- and intermediate-time 1D dynamics	local gates or projected evolution	time step, truncation, $\chi$, conservation, reachable time
MPS on cylinders	quasi-1D views of 2D systems	mature 1D machinery and controlled long direction	width, length, ordering, $\chi$, edge and sector effects
PEPS	2D ground-state structure	lattice-matched area-law capacity	state $D$, environment cutoff, update, contraction scheme
tree tensor network	hierarchical or cluster geometry	loop-free contraction and multiscale partitions	tree choice, bond profile, translation artifacts
MERA	scale-resolved and critical states	bounded causal cones and explicit scale layers	bond dimension, layers, architecture, scaling-operator convergence
MPO or purification	operators and thermal states	structured representation of mixed-state tasks	operator bond, ancilla choice, positivity, temperature step

No row is a guarantee. A method is a candidate only after the target, geometry, and observable are declared.

From area law to algorithm

For an MPS cut with bond dimension $\chi$,

$$\operatorname{rank}\rho_A \le \chi, \qquad S_A \le \ln\chi.$$

This is a capacity statement. It does not say that:

• every area-law state has a small useful $\chi$ at the desired precision;
• the Schmidt tail decays rapidly enough for the observable of interest;
• an optimizer will find the best finite-$\chi$ state;
• a higher-dimensional network is easy to contract;
• a finite-$\chi$ result has reached the physical correlation length.

On a cylinder of width $W$, an area law of the form

$$S_{\mathrm{cut}} \sim \alpha W$$

implies the capacity requirement

$$\chi \gtrsim e^{\alpha W}$$

for a chain-ordered MPS crossing that cut. The MPS cost can therefore grow exponentially with cylinder width even when the two-dimensional state obeys an area law.

PEPS changes the virtual geometry: a two-dimensional boundary cuts many bonds, so fixed local $D$ can supply boundary-law capacity. That representational advantage comes with a harder contraction problem. MERA distributes bonds by scale, which favors critical structure but introduces a more constrained architecture and optimization.

The correct implication is:

$$\begin{gathered} \text{favorable entanglement structure} \\ \text{may enable a compact ansatz}, \\ \text{compactness} \not\Longrightarrow \text{easy evaluation}, \\ \text{compactness} \not\Longrightarrow \text{easy optimization}. \end{gathered}$$

Bond dimension is not an error bar

The symbols $\chi$ and $D$ are resource controls, not observables.

Their meaning depends on:

• network topology;
• physical site grouping and ordering;
• symmetry-resolved block structure;
• open, periodic, finite, or infinite geometry;
• state versus operator representation;
• local physical dimension;
• the contraction and optimization method.

Two calculations at “bond dimension 256” need not have comparable state capacity or cost.

Use observable convergence

For each target observable, compare successive controls:

$$\Delta_O(D_2,D_1) = \left| O(D_2) - O(D_1) \right|.$$

A small $\Delta_O$ is useful only if:

• both calculations are independently optimized;
• environment errors are smaller than the change;
• size and boundary effects are separately controlled;
• the sequence is not trapped on one metastable branch;
• the observable is not artificially pinned by a symmetry or unit-cell choice.

Use state diagnostics

For a normalized candidate eigenstate,

$$\sigma_H^2 = \langle H^2\rangle - \langle H\rangle^2.$$

An exact eigenstate has zero variance. A decreasing variance is stronger evidence than energy stationarity alone, though fidelity conclusions still require spectral information.

Schmidt values, truncation weights, canonical residuals, transfer spectra, and symmetry charges diagnose other aspects of the state. No single scalar replaces the full evidence ledger.

Convergence designs by regime

Finite gapped chain

A defensible finite-chain study varies:

$$\begin{gathered} L, \qquad \chi, \\ \text{sweeps or iterations}, \\ \text{initialization}. \end{gathered}$$

It checks energy variance, central versus edge observables, symmetry sector, representative Schmidt tails, and agreement with exact diagonalization at smaller $L$.

Open boundaries often reduce MPS cost, but they introduce edge profiles. Bulk claims should use a central window or a controlled boundary extrapolation.

Critical chain

At a critical point, finite size and finite bond dimension produce competing infrared cutoffs:

$$\xi_{\mathrm{eff}} \sim \min \left( L, \xi_\chi \right).$$

Increasing $L$ at fixed $\chi$ can reveal only the artificial finite-entanglement plateau. A critical study therefore needs a two-dimensional grid in $(L,\chi)$ or a justified finite-entanglement scaling protocol.

Cylinder or ladder

For a cylinder, vary width and length separately:

$$W \longrightarrow \infty, \qquad L \longrightarrow \infty.$$

Also vary the one-dimensional path through the sites, boundary pinning, flux or topological sector, and bond dimension. A stable result on one width is a quasi-one-dimensional result, not automatically a two-dimensional thermodynamic conclusion.

PEPS calculation

At minimum, vary both:

$$D \qquad \text{and} \qquad \chi_{\mathrm{env}},$$

where $\chi_{\mathrm{env}}$ denotes the boundary or environment control used by the contraction scheme.

Convergence in $D$ at one fixed $\chi_{\mathrm{env}}$ can merely converge the bias of an under-resolved environment. Compare contraction schemes or environment constructions when the claimed accuracy is close to their differences.

Real-time evolution

Real-time tensor-network evolution has at least three horizons:

$$\delta t, \qquad \chi, \qquad t_{\max}.$$

Decrease the time step, increase the retained bond dimension or decrease the truncation threshold, and monitor conservation laws. Entanglement growth can make the reachable time increase only slowly with computational effort.

A smooth curve beyond the entanglement horizon is not evidence of accuracy.

Thermal and mixed-state calculations

Specify whether the density operator is represented as an MPO, a purification, a minimally entangled ensemble, or another construction. Control:

• inverse-temperature step;
• operator or purification bond dimension;
• ancilla gauge and normalization;
• positivity or complete-positivity assumptions when relevant;
• spatial size and thermal correlation length;
• observable-specific contraction error.

Error ledger

For an estimated thermodynamic observable,

$$\begin{aligned} \widehat O - O_\infty &= \delta_{\mathrm{basis}} + \delta_{\mathrm{size}} + \delta_{\mathrm{bond}} \\ &\quad + \delta_{\mathrm{contract}} + \delta_{\mathrm{opt}} \\ &\quad + \delta_{\mathrm{step}} + \delta_{\mathrm{other}}. \end{aligned}$$

These terms need not be independent or additive in a strict probabilistic sense. The decomposition is an audit tool: every active approximation needs a named control.

Error source	Typical control	Typical false conclusion
onsite basis	enlarge local cutoff or orbital set	apparent convergence in a truncated model
finite size	increase length and compare boundaries	crossover or edge effect called a phase
finite width	compare cylinders or ladders	quasi-1D behavior called 2D bulk physics
finite bond	increase $\chi$ or $D$	artificial entropy or correlation-length saturation
environment	increase $\chi_{\mathrm{env}}$ and compare contractions	biased PEPS energy called variationally converged
optimization	restarts, sectors, unit cells, residuals	metastable tensor called the ground state
time or temperature step	reduce step and change integrator	discretization drift called dynamics
observable extraction	independent contractions and sum rules	normalization or insertion error called physics

A reliable workflow

1. Declare the physical target. State $H$, geometry, boundary, sector, state or ensemble, observable, and limit.
2. Choose the canonical representation. Name the network, ordering, physical grouping, symmetries, and bond profile.
3. List every approximation. Include local cutoff, bond dimension, environment, optimizer, step size, and finite geometry.
4. Validate local algebra. Check tensor shapes, charges, Hermiticity, canonical or isometric residuals, and normalization.
5. Benchmark a tractable overlap window. Use exact diagonalization, analytic limits, or another method.
6. Run a convergence grid. Do not vary only the cheapest control.
7. Track state and observable diagnostics. Include variance, Schmidt data, correlations, and symmetry labels as appropriate.
8. Separate numerical and physical extrapolations. First control the finite calculation; then infer the intended limit.
9. Report the evidence horizon. State where bond growth, width, time, or contraction uncertainty prevents a stronger conclusion.

Choosing stopping criteria

Stopping because the code no longer changes visibly is not a scientific criterion.

A defensible stopping rule ties numerical changes to the target:

$$\left| O_{k+1}-O_k \right| < \epsilon_O$$

over a declared sequence of controls, with independent evidence that omitted errors are below $\epsilon_O$ or explicitly included in the uncertainty.

For a phase classification, stopping criteria may require more than one local observable:

• gap or correlation-length behavior;
• symmetry quantum numbers and order parameters;
• entanglement spectrum or topological-sector diagnostics;
• stability across boundary, initialization, and unit-cell choices;
• finite-size or finite-entanglement scaling.

The evidence should match the claim.

Reporting checklist

A reusable tensor-network result should record:

• Hamiltonian, couplings, units, and constant shifts;
• physical geometry, boundaries, and site or orbital ordering;
• network family and finite, infinite, or periodic form;
• local dimensions and all bond dimensions;
• exact symmetries and target sector;
• contraction algorithm and environment controls;
• optimization update, solver tolerance, sweep count, restarts, and initialization;
• truncation rule and any discarded-weight summaries;
• time or inverse-temperature integrator and step;
• convergence tables for every active axis;
• variance, residual, normalization, and symmetry checks;
• exact or cross-method benchmarks;
• observable extraction and uncertainty procedure;
• hardware and software provenance when performance or reproducibility matters.

Common mistakes

• Treating the canonical tensor-network article and a computational workflow as interchangeable.
• Choosing a network by dimension alone rather than by target state, geometry, and observable.
• Reversing an area-law implication and assuming every area-law state is easy.
• Calling bond dimension an accuracy.
• Comparing $\chi$ or $D$ across different topologies as if they were the same resource.
• Varying state bond dimension while fixing an under-resolved environment.
• Declaring DMRG convergence from the final sweep’s energy change alone.
• Reporting only the maximum local discarded weight as a global state error.
• Ignoring orbital or site ordering in an MPS calculation.
• Using one unit cell when translation breaking is a live possibility.
• Interpreting a finite-$\chi$ transfer correlation length as a Hamiltonian gap.
• Treating a finite cylinder as the two-dimensional thermodynamic limit.
• Extending a real-time curve beyond its entanglement-controlled horizon.
• Assuming a low variational energy certifies long-range or topological observables.

Exercises

Exercise 1: Classify the four gates

Assign each observation to representability, evaluability, searchability, or inferential control.

1. A PEPS energy changes when the environment cutoff is doubled.
2. Different random initial tensors converge to different energies at the same $D$.
3. Exact diagonalization shows that no state at the tested $\chi$ reproduces the target Schmidt rank.
4. A converged finite cylinder changes phase indicator when its width increases.

Solution

1. This is an evaluability problem: the approximate contraction has not converged.
2. This is a searchability problem: optimization depends on initialization.
3. This is a representability problem at the tested bond dimension.
4. This is an inferential-control problem: the finite-width result does not yet support the intended dimensional limit.

The gates can interact, but naming the first failed gate identifies the next diagnostic.

Exercise 2: Cylinder-width cost

Suppose the entropy across a cylinder cut behaves as

$$S = \alpha W, \qquad \alpha = \frac{1}{2}.$$

Use $S\le\ln\chi$ to estimate the minimum capacity at widths $W=8$ and $W=12$, and find their ratio.

Solution

The capacity condition gives

$$\chi \ge e^{\alpha W}.$$

At $W=8$,

$$\chi_8 \gtrsim e^4 \approx 55.$$

At $W=12$,

$$\chi_{12} \gtrsim e^6 \approx 403.$$

The ratio is

$$\frac{\chi_{12}}{\chi_8} \gtrsim e^2 \approx 7.4.$$

These are capacity lower bounds, not accuracy predictions. Schmidt-tail structure and the desired observables can require larger bonds.

Exercise 3: Diagnose false PEPS convergence

A study reports energies at $D=2,3,4,5$ using one fixed environment cutoff. The final two energies agree to eight digits, so the authors claim convergence in $D$. What evidence is missing?

Solution

The contraction error is not controlled. A fixed, insufficient environment can make every state energy approach the same biased value.

At each relevant $D$, the study should increase the environment cutoff until the energy and target observables stabilize. It should also compare environment constructions when their systematic differences matter, verify normalization, and check whether the energy evaluation preserves the claimed variational property.

Only after contraction error is below the observed $D$ dependence does the sequence test finite-bond convergence.

Exercise 4: Design a critical-chain grid

A finite-MPS calculation targets a critical exponent. Explain why a sequence with increasing $L$ at fixed $\chi$ is insufficient and propose a minimal convergence design.

Solution

At fixed $\chi$, the MPS develops a finite-entanglement correlation length $\xi_\chi$. Once $L\gg\xi_\chi$, increasing $L$ probes the finite-$\chi$ state rather than the scale-free target.

A minimal design uses several $L$ values at each of several increasing $\chi$ values. It records $\xi_\chi$, energy variance, fitted observables, and the scaling window. The analysis should show either:

• a finite-size regime with $L\ll\xi_\chi$ that is stable as $\chi$ increases; or
• a justified finite-entanglement scaling collapse with stable exponents.

Unit cell, boundaries, symmetry sector, and initialization should also be tested when they can affect the critical state.

Exercise 5: Audit an efficiency claim

A paper states: “Our two-dimensional state has an exact PEPS with $D=4$, so all local observables can be computed efficiently and exactly.” Identify the valid part and the unsupported conclusion.

Solution

The valid statement is representational: an exact $D=4$ PEPS is a compact description of the state in the declared geometry.

The unsupported step is from compact representation to efficient exact contraction. A PEPS norm or local expectation value forms a looped two-dimensional contraction, and exact contraction can remain computationally hard at fixed $D$. The paper must specify a contraction algorithm, its environment controls, and an error analysis.

Special isometric, free, stabilizer, or otherwise structured PEPS may be exactly tractable, but that requires an additional theorem beyond finite bond dimension.

References

• S. R. White, “Density Matrix Formulation for Quantum Renormalization Groups,” Physical Review Letters 69, 2863–2866 (1992). doi:10.1103/PhysRevLett.69.2863
• G. Vidal, “Efficient Classical Simulation of Slightly Entangled Quantum Computations,” Physical Review Letters 91, 147902 (2003). doi:10.1103/PhysRevLett.91.147902
• F. Verstraete and J. I. Cirac, “Matrix Product States Represent Ground States Faithfully,” Physical Review B 73, 094423 (2006). doi:10.1103/PhysRevB.73.094423
• M. B. Hastings, “An Area Law for One-Dimensional Quantum Systems,” Journal of Statistical Mechanics: Theory and Experiment 2007, P08024 (2007). doi:10.1088/1742-5468/2007/08/P08024
• N. Schuch, M. M. Wolf, F. Verstraete, and J. I. Cirac, “Computational Complexity of Projected Entangled Pair States,” Physical Review Letters 98, 140506 (2007). doi:10.1103/PhysRevLett.98.140506
• G. Vidal, “Entanglement Renormalization,” Physical Review Letters 99, 220405 (2007). doi:10.1103/PhysRevLett.99.220405
• F. Verstraete, V. Murg, and J. I. Cirac, “Matrix Product States, Projected Entangled Pair States, and Variational Renormalization Group Methods for Quantum Spin Systems,” Advances in Physics 57, 143–224 (2008). doi:10.1080/14789940801912366
• U. Schollwöck, “The Density-Matrix Renormalization Group in the Age of Matrix Product States,” Annals of Physics 326, 96–192 (2011). doi:10.1016/j.aop.2010.09.012
• J. Haegeman, C. Lubich, I. Oseledets, B. Vandereycken, and F. Verstraete, “Unifying Time Evolution and Optimization with Matrix Product States,” Physical Review B 94, 165116 (2016). doi:10.1103/PhysRevB.94.165116
• M. C. Bañuls, “Tensor Network Algorithms: A Route Map,” Annual Review of Condensed Matter Physics 14, 173–191 (2023). doi:10.1146/annurev-conmatphys-040721-022705

Further study

• Tensor Networks Preview for canonical definitions, graph bounds, MPS, PEPS, MERA, and network-level error accounting.
• Matrix Product States Preview for canonical gauges, transfer operators, Schmidt truncation, and one-dimensional structure.
• Area Laws for theorem status and compression caveats.
• Entanglement Spectrum for Schmidt tails and truncation diagnostics.
• Computational Many-Body Overview for cross-method selection and validation.
• DMRG Preview for finite-system MPS optimization, sweeps, truncation, and convergence evidence.
• Finite-Size Scaling in Numerics for thermodynamic and critical extrapolation.
• Boundary Conditions on Lattices for open, periodic, cylinder, and twist choices.
• Time-Dependent Variational Principle for projected dynamics on variational manifolds.
• Validation Tests for independent computational checks.