Continuum Functional Equations and Controlled Truncations

Continuum functional methods rewrite QFT as exact identities or flows for correlation functions and effective actions. Their practical power comes from finite closures; their principal risk is that a precisely solved closure can be mistaken for a precisely solved theory. This chapter develops a common inference sequence: exact equation, declared external inputs, retained and omitted structures, symmetry and renormalization tests, branch and numerical solution, analytic interpretation, independent benchmark, and separated errors.

Helpful background. Interacting fields and effective descriptions supplies the observable and limiting framework, while Schwinger–Dyson identities supplies the basic functional identity.

Enter this chapter

The nine pages answer three layers of question:

1. Construction: how are hierarchies, 2PI actions, bound-state equations, and functional flows obtained?
2. Closure: which functions and tensors are retained, how are symmetries and counterterms preserved, and which nonlinear branch is selected?
3. Interpretation: how do Euclidean solutions become spectra or observables, and which errors and external benchmarks support the claim?

All Euclidean formulas declare their continuation convention before a Lorentzian pole is inferred. Gauge-fixed functions are kept distinct from gauge-invariant observables. Solver residual, truncation spread, branch ambiguity, symmetry violation, analytic-continuation dependence, parameter covariance, and numerical error are never treated as interchangeable.

Choose the route by the problem

Reading routes through continuum functional equations and their validation.

Question	Start here	Main result	Required limitation
How does an exact identity become an infinite hierarchy?	Schwinger–Dyson hierarchies and inputs	Renormalized scalar two-point equation, missing four-point input, zero-dimensional fixture, and validation record	Exact hierarchy is not a finite solution
How should a hierarchy be closed?	Closure, symmetry, and branch selection	Retained and omitted tensor basis, Ward-identity example, branch tests, and separated errors	Solver convergence is not truncation control
How do self-consistent skeleton equations avoid double counting?	2PI and nPI effective actions	Double Legendre transform, stationary Dyson equation, and Hartree-plus-sunset factors	Conserving does not imply every Ward or crossing identity
How is a relativistic bound state extracted?	Bethe–Salpeter and Faddeev equations	Pole equation, canonical normalization, and kernel–self-energy relation	A Euclidean eigenvalue is not automatically a physical pole
How is the exact functional-RG flow used?	Functional-RG applications and cross-checks	O(N) effective-potential projection and regulator/order variation	The exact flow becomes approximate after finite projection
How are propagators, vertices, and spectra solved together?	Coupled functional systems	Dependency ordering, coherent kernel variation, joint residuals, and covariance	Separately fitted components can be mutually inconsistent
What licenses a spectral or real-time claim?	Complex momentum, spectra, and real time	Direct complex solution, spectral representation, and inverse-problem tests	Finite Euclidean data do not determine a unique spectrum without assumptions
Which checks support the reported digits?	Validation and error control	Minimum validation battery and six separately reported error channels	Related closures are not independent evidence
What changes in a gauge-fixed system?	BRST constraints and the Gribov problem	Landau-gauge propagator–vertex tests, copy sensitivity, and observable boundary	Gauge-fixed positivity violation alone does not prove confinement

A coherent reading path

Begin with Schwinger–Dyson Hierarchies and Renormalization Inputs. Its exact zero-dimensional fixture makes the hierarchy, closure, positivity rejection, held-out residual, and external benchmark fully explicit. Then read Closure, Symmetry Constraints, and Branch Selection to learn the finite-problem record.

Choose a construction:

• 2PI and nPI Effective Actions for variational skeleton equations;
• Bethe–Salpeter and Faddeev Bound-State Equations for pole residues and symmetry-related kernels;
• Functional-RG Applications and Cross-Checks for scale-dependent effective actions.

Use Coupled Propagator, Vertex, and Bound-State Systems before combining those components. Add Complex-Momentum, Spectral, and Real-Time Information when the desired observable lies outside the Euclidean domain. Finish with Functional-Method Validation and Error Control, and use Gauge Fixing, BRST Constraints, and the Gribov Problem when the unknowns are gauge dependent.

Shared control contract

Every reader-facing conclusion should have the form

$$\begin{aligned} &\text{exact identity or flow} +\text{external inputs} +\text{finite closure} \\ &\quad +\text{branch and numerical solution} +\text{independent validation} \longrightarrow \text{bounded observable claim}. \end{aligned}$$

“External inputs” includes regulator, counterterms, finite renormalization conditions, boundary or vacuum sector, gauge prescription, and positivity class. “Finite closure” includes omitted functions and tensor structures, not only retained equations. “Independent validation” must not reuse the decisive ansatz without recording the covariance.

The common error vector is

$$\Delta =\left( \Delta_{\mathrm{trunc}}, \Delta_{\mathrm{branch}}, \Delta_{\mathrm{sym}}, \Delta_{\mathrm{cont}}, \Delta_{\mathrm{param}}, \Delta_{\mathrm{num}} \right).$$

It is more informative than one undocumented error bar. Different components may be bounds, discrete alternatives, or statistical covariances; they should not be combined mechanically.

What exact means here

• A Schwinger–Dyson identity is exact for the regulated measure; closing its hierarchy is additional.
• A 2PI representation is exact before skeleton truncation; a finite-loop stationary point is additional.
• A Bethe–Salpeter pole equation is exact with the exact kernel; modeling that kernel and continuing Euclidean momenta are additional.
• The Wetterich equation is exact in full theory space; a finite ansatz and projection are additional.
• Euclidean reconstruction is exact with complete axiomatic data under the reconstruction hypotheses of Osterwalder and Schrader 1973, pp. 86–94; inversion from finite uncertain samples needs extra analytic information.
• BRST and Slavnov–Taylor identities control a declared gauge-fixed measure; they do not solve global Gribov ambiguity or turn gauge-fixed correlators into gauge-invariant observables.

The distinction between exact construction and finite deployment is central to the original 2PI formulation Cornwall, Jackiw, and Tomboulis 1974, §§ II–III and the exact functional flow Wetterich 1993, Eqs. (1)–(7).

Review the chapter

1. A truncated equation is solved with residual $10^{-12}$, but the next tensor basis shifts the observable by $4\%$. Which precision is justified?

Solution

The equation is solved accurately inside the original closure, but the observed truncation spread is $4\%$. Unless a higher-order sequence reduces that spread predictably, the scientific result is only controlled at roughly the percent level, not twelve numerical digits.

2. Why must a Bethe–Salpeter kernel change when a shared self-energy ansatz is varied?

Solution

In a symmetry-preserving construction the kernel is related by $K=-\delta\Sigma/\delta S$. Changing the ansatz changes both the constituent dressing and its functional derivative. Freezing one side can violate the Ward identity and discards their covariance.

3. A gauge-fixed gluon propagator has no positive spectral density. What conclusion is licensed?

Solution

The gauge-fixed field does not reconstruct a positive-metric asymptotic one-particle state under that spectral test. This is not by itself a gauge-invariant confinement result. The gauge, copy prescription, truncation, and a separate gauge-invariant observable must be stated.

References

• Cornwall, John M., R. Jackiw, and E. Tomboulis. “Effective Action for Composite Operators.” Physical Review D 10 (1974): 2428–2445. DOI.
• Osterwalder, Konrad, and Robert Schrader. “Axioms for Euclidean Green’s Functions.” Communications in Mathematical Physics 31 (1973): 83–112. DOI.
• Wetterich, Christof. “Exact Evolution Equation for the Effective Potential.” Physics Letters B 301 (1993): 90–94. DOI.