Complexity and Continuum Resource Costs

Complexity is not a property of a QFT state or operator in isolation. It is the minimum cost of a declared task: one must specify the reference and target, the admissible transformations, the error criterion, the cost assigned to those transformations, and the regulator or operational resource that makes the minimization meaningful. This chapter develops that discipline for circuits, geometric costs, Gaussian and interacting fields, mixed states, Euclidean preparation, Krylov growth, symmetry constraints, and continuum comparisons.

Helpful background. Why Continuum QFT Does Not Factorize Naively explains why continuum subregions cannot automatically be treated as qubit registers. Direct Sums, Tensor Products, and Index Structure supplies the finite-regulator tensor language. Basis Dependence, Computational Complexity, and No Universal Cure shows how a change of representation can change an algorithmic cost, while Operator Spreading and Scrambling supplies the dynamics that Krylov measures summarize. These are useful entry routes, not conditions for reading the overview.

A complexity statement is a complete task tuple

Write a proposed complexity as

$$\mathfrak C_{\mathcal T} =\inf_{P\in\mathcal P(\mathcal T)}F[P], \qquad \mathcal T=(X_{\rm ref},X_{\rm tar},\mathcal A,d,\epsilon,F,\Lambda,\mathcal R).$$

Here $X$ says whether the object is a state, unitary, channel, operator, function, or definable domain; $\mathcal A$ is the admissible operation or description class; $d\leq\epsilon$ defines success; $F$ is the cost; $\Lambda$ denotes the regulator and continuum prescription; and $\mathcal R$ lists the physical, computational, or descriptive resource being counted. Two numerical values are comparable only after a translation preserves these fields or proves controlled inequalities between them.

This definition separates four families that are often conflated:

• circuit or control cost minimizes over allowed transformations that prepare or implement a target Nielsen 2006, §§2–4; regulated QFT examples make the reference and ultraviolet dependence explicit Jefferson and Myers 2017, §§2–4;
• Krylov complexity is a moment of an operator wavefunction in a Lanczos basis fixed by a generator and inner product Parker et al. 2019, §§II–IV;
• computational complexity counts time, queries, samples, memory, or another algorithmic resource for a promise problem;
• format and degree describe the logical-geometric specification of definable functions or domains in a chosen sharp o-minimal structure Binyamini, Novikov, and Zack 2022, §§1–3.

None is a universal surrogate for the others. An exact formula can have low descriptive format but high physical preparation cost; a rapidly spreading operator can have large Krylov complexity while a particular target state remains easy to prepare.

Choose a route

Goal	Route	Stop when you can…
Formulate a claim	What Task Does Complexity Answer? → Circuit Complexity in Quantum Field Theory	write every field of $\mathcal T$ and identify the counted resource
Use geometric methods	circuit complexity → Cost Geometry, Gate Sets, and Reference States → Complexity of Bosonic and Fermionic Gaussian States	distinguish a geodesic extremum from the globally least-cost path
Compare object types	task definition → State, Unitary, Channel, and Operator Complexity → Mixed-State, Purification, and Formation Complexity	state which freedoms are quotiented or optimized over
Leave the Gaussian sector	Gaussian states → Complexity Beyond Gaussian and Free Fields	attach a truncation, renormalization, and error estimate to the result
Treat Euclidean preparation	circuit complexity → Path-Integral and Euclidean Preparation Complexity	separate a preparation protocol from a coordinate-dependent action-like proposal
Study operator growth	task definition → Krylov Complexity and Operator Growth	specify the seed, inner product, Liouvillian, and truncation
Enforce physical restrictions	circuit complexity → Symmetry, Gauge Constraints, and Local Gate Sets → Operational Preparation Cost and Energy-Constrained Bounds	identify which gates are admissible and which controls are physically bounded
Seek a continuum comparison	Regulator Dependence and Continuum Complexity → Complexity, Chaos, and Computational Claims	show what survives matched regulators and independent chaos controls

Dictionary from objects to optimization problems

The first figure asks the question that should precede every calculation: what is being optimized, over what admissible set, and with which equivalences?

A complexity value is defined only after the target object selects an admissible family of paths or descriptions. Circuit length, physical control cost, Krylov spread, algorithmic resources, and sharp o-minimal format or degree answer different questions. The diagram is schematic and not to scale.

The semantic comparison is:

Complexity proposals compared by target task, admissible class, cost, reference, tolerance, regulator, continuum scaling, operational interpretation, and residual ambiguity.

Proposal	Target task	Admissible class and cost	Reference and tolerance	Regulator and continuum question	Operational or invariant content	Principal ambiguity
State circuit	Prepare ρtarget from ρreference	Declared local or Gaussian gates; size, depth, or path length	State reference; trace, fidelity, or covariance error	Fixed physical error as lattice spacing tends to zero	Preparation cost for the chosen gate model	Reference, stabilizer quotient, and gate penalties
Unitary circuit	Synthesize a unitary on a declared domain	Gates modulo permitted phases; size or depth	Identity or reference unitary; channel error	Energy-constrained domain and regulator matching	Implementation resource on that domain	Precision, generator normalization, and nonlocal gates
Channel complexity	Implement a completely positive channel	Allowed dilations, ancillas, and controls; circuit or control cost	Reference channel and channel distance	Energy constraint and environment model	Implementation resource after declared minimization	Dilation freedom and uncharged environment preparation
Krylov complexity	Evolve one seed operator in a Lanczos basis	Fixed Liouvillian and inner product; first Krylov-index moment	Seed and inner product fix the basis	Ultraviolet spectral tail and order of limits	Basis-relative operator spread	Seed, thermal inner product, and truncation
Computational complexity	Estimate or simulate an observable for an instance family	Algorithms satisfying a promise; time, queries, samples, or memory	Input encoding, error, and success probability	Family of regulated instances and scaling variable	Invariant only under stated reductions	Encoding and computational model
Sharp format and degree	Specify a definable set or function	Formulas in a fixed structure; filtration pair (F, D)	Presentation and reduction rules	Not a state-preparation continuum limit	Descriptive-geometric bounds	Choice of structure and presentation

Where comparisons fail

The second figure groups the most common hidden changes of task. A quoted scaling is not robust until these changes have been tested or explicitly excluded.

Reference sensitivity, gate nonuniqueness, regulator dependence, symmetry constraints, and unbounded controls are distinct failure modes. A link from complexity growth to chaos or computational hardness requires separate evidence after those controls. The diagram is schematic.

The thirteen pages

1. What Task Does Complexity Answer? supplies the comparison tuple and the four-way distinction among resource notions.
2. Circuit Complexity in Quantum Field Theory defines regulated approximation by admissible circuits.
3. Cost Geometry, Gate Sets, and Reference States develops Nielsen geometry and its normalization choices.
4. State, Unitary, Channel, and Operator Complexity derives the inequivalent quotient freedoms of four targets.
5. Gaussian-State Complexity turns covariance data into controlled solvable examples.
6. Mixed-State, Purification, and Formation Complexity compares optimizations over spectra, bases, and ancillas.
7. Interacting-Field Complexity states what perturbative and variational estimates actually control.
8. Path-Integral and Euclidean Preparation Complexity separates Euclidean protocols from proposal-specific costs.
9. Krylov and Operator-Growth Complexity derives the Lanczos-chain probability distribution.
10. Complexity with Symmetry, Gauge, and Locality Constraints restricts admissible transformations before minimizing.
11. Regulator Dependence and Continuum Complexity tests divergent terms and finite remainders across matched schemes.
12. Operational Preparation Cost and Energy-Constrained Bounds connects abstract paths to bounded physical controls.
13. Complexity, Chaos, and Computational Claims gives a claim matrix and explicit nonimplications.

Conventions and scope boundaries

Circuit products act in their displayed time order; a page states whether its geometric convention is left- or right-invariant. Bosonic covariance matrices use canonical quadratures with the commutator normalization declared locally, while fermionic pages declare Majorana normalization. A cost carries its generator normalization and penalty schedule. Continuum claims compare a family of regulated tasks at fixed physical tolerance; subtracting a divergence without a permitted local cost and a matched reference does not create a universal observable.

Algorithms and tensor-network costs are developed in Volume 8, while physical operator dynamics and chaos diagnostics are developed in Volume 11. Holographic complexity proposals belong to Volume 15 and do not define generic QFT complexity here. A reproducible verification should compare matched finite-mode examples with identical regulators and cost conventions.

Review the chapter

Task translation. Two papers report different ultraviolet exponents for “the complexity of the vacuum.” Give a minimum comparison test.

Verification criteria

Match the target and reference families, gate generators and their normalization, locality and penalty rules, cost functional, error metric and tolerance, spatial volume, zero-mode treatment, regulator, and order of limits. If no controlled translation preserves these data, the exponents answer different questions.

Operational meaning. A geometric path has length $L$ and can be traversed in arbitrarily small parameter time. Why is $L$ not yet a laboratory duration?

Verification criteria

The path parameter is not physical time. A duration bound needs a control Hamiltonian, amplitude or energy limits, locality, bandwidth, and an implementation map from geometric generators to controls. Without those data, reparameterization changes duration while leaving $L$ fixed.

Nonimplication. Does rapidly growing Krylov complexity prove chaos?

Verification criteria

No. It proves spreading in the Lanczos basis fixed by a seed, inner product, and generator. A chaos claim also needs independent dynamical or spectral diagnostics, symmetry resolution, finite-size and regulator controls, and an argument excluding integrable or free counterexamples.

References

• Binyamini, Gal, Dmitri Novikov, and Benny Zack. “Sharply o-Minimal Structures and Sharp Cellular Decomposition.” arXiv:2209.10972 (2022; revised 2024). Preprint.
• Jefferson, Ro, and Robert C. Myers. “Circuit Complexity in Quantum Field Theory.” Journal of High Energy Physics 10 (2017): 107. DOI. Open PDF.
• Nielsen, Michael A. “A Geometric Approach to Quantum Circuit Lower Bounds.” Quantum Information & Computation 6 (2006): 213–262. Open PDF.
• Parker, Daniel E., Xiangyu Cao, Alexander Avdoshkin, Thomas Scaffidi, and Ehud Altman. “A Universal Operator Growth Hypothesis.” Physical Review X 9 (2019): 041017. DOI. Open PDF.