Theorem-First Claim Records: Objects, Hypotheses, Conclusions, and Status

A statement about a quantum field theory becomes mathematically assessable only after its objects, quantifiers, hypotheses, and conclusion have been fixed. The same sentence—“the continuum theory exists,” for example—can mean convergence of finitely many correlators, construction of a probability measure, reconstruction of a Hilbert-space theory, or existence of a non-Gaussian interacting model. Those are different claims. A theorem-first record preserves the difference and states exactly what the available argument proves.

Required background. Limits, Completeness, and Modes of Convergence supplies the topologies, modes of convergence, and order-of-limits language used below. Helpful background. Claim Status, Freshness, and Research Handoffs treats dated scientific claims; Graded Spacetime Symmetry and the Supersymmetry Theorems shows why hypotheses control a classification theorem; Replica and Entropy Calculation Verification separates a calculation from its continuum interpretation; and Fixed-Background, Semiclassical, Gravitational-EFT, and Quantum-Gravity Regimes illustrates regime-dependent conclusions.

The anatomy of a determinate QFT claim

A useful record can be represented by the tuple

$$\mathcal C=(\mathsf O,\mathsf V,\mathsf D,\mathsf Q,\mathsf H,\mathsf M,\mathsf K,\mathsf U,\mathsf N,\mathsf S,\mathsf t).$$

The symbols are not extra formalism for its own sake. Each one answers a question that can change the truth value of the statement.

Field	Question that must be answered
$\mathsf O$: objects	Are the primitives random distributions, operator-valued distributions, a net of algebras, states, functors, or a formal power series?
$\mathsf V$: variables	Which spacetime dimension, masses, couplings, boundary conditions, regulators, and observables vary?
$\mathsf D$: domains	Which spacetime, test-function space, operator domain, support class, and regularity class are used?
$\mathsf Q$: quantifiers	Is the assertion for every coupling, for sufficiently small coupling, for a subsequence, or merely for one computed parameter set? In which order are limits taken?
$\mathsf H$: hypotheses	Which positivity, covariance, locality, spectral, compactness, stability, or uniform-bound assumptions are imposed?
$\mathsf M$: map or construction	What map produces the claimed object—weak limit, reconstruction, completion, GNS representation, renormalized extension, or comparison functor?
$\mathsf K$: conclusion	Is the result existence, uniqueness, equivalence, an error estimate, a structural property, or an obstruction?
$\mathsf U$: uniqueness strength	Unique literally, up to isomorphism, up to unitary equivalence, locally quasiequivalent, or dependent on choices?
$\mathsf N$: nonconverse	Does the reverse implication hold? If not, what counterexample or missing hypothesis prevents it?
$\mathsf S,t$: source and date	Which theorem version, correction, and inspection date support the wording?

The record also names the mode of justification. An axiom stipulates a property. A construction supplies an object and proves stated properties. A reconstruction theorem starts from one class of data and produces another. A formal perturbative expansion is coefficientwise, generally in a ring such as $\mathbb C[[\lambda]]$, and does not by itself define a function at nonzero $\lambda$. A numerical result concerns its finite inputs, algorithm, precision, and error controls. A conjecture, obstruction, and open problem have still different logical forms. None should be silently relabeled as another.

This grammar makes the conclusion read like a typed implication,

$$\forall x\in\mathsf D,\qquad \mathsf H(x)\Longrightarrow \mathsf K\bigl(\mathsf M(x)\bigr),$$

with every symbol given a domain. If the theorem instead asserts existence of a subsequence, the quantifiers must say so. If a choice of gauge, renormalization prescription, state, or completion enters $\mathsf M$, the result records whether $\mathsf K$ is invariant under that choice.

Quantifier order is part of the theorem

For a regulated family $X_{a,L}$, the expressions

$$\lim_{a\downarrow0}\lim_{L\to\infty}X_{a,L}, \qquad \lim_{L\to\infty}\lim_{a\downarrow0}X_{a,L}, \qquad \lim_{\substack{a\downarrow0\\L=L(a)\to\infty}}X_{a,L}$$

are three distinct claims. Equality requires a theorem, usually based on uniform estimates or a joint compactness-and-uniqueness argument. Likewise,

$$\forall f\in\mathcal S(\mathbb R^d),\quad X_n(f)\Rightarrow X(f)$$

does not automatically assert convergence in law of $X_n$ as random elements of $\mathcal S'(\mathbb R^d)$. The latter also needs tightness in a specified topology and identification of every subsequential limit. A claim about moments needs moment determinacy or a separate generating-functional argument before it becomes a claim about measures.

The same discipline applies to error estimates. “The approximation becomes exact” should be replaced by a bound such as

$$\sup_{x\in K}\lVert F_N(x)-F(x)\rVert\leq C_K N^{-\alpha},$$

where the compact set $K$, norm, exponent, and parameter dependence of $C_K$ are stated. Without them, the advertised uniformity may be absent.

Four-dimensional lattice φ⁴: four separate obligations

Consider a real field $\phi_x$ on a finite lattice $\Lambda_{a,L}\subset a\mathbb Z^4$ with a stable finite-volume probability measure

$$d\mu_{a,L}(\phi)=Z_{a,L}^{-1} \exp\!\left[-a^4\sum_{x\in\Lambda_{a,L}} \left(\frac12(\nabla_a\phi)_x^2+\frac12m_0^2(a)\phi_x^2 +\lambda_0(a)\phi_x^4\right)\right]\prod_x d\phi_x .$$

The slogan “four-dimensional $\phi^4$ exists in the continuum” decomposes as follows.

1. Finite-cutoff existence. For fixed $a>0$ and finite $L$, is $Z_{a,L}$ finite and are the selected moments defined? Stability of the quartic term answers only this finite-dimensional question.

2. A tuned limiting family. Which functions $m_0^2(a)$ and $\lambda_0(a)$ are chosen? Is $L\to\infty$ taken first, jointly with $a\downarrow0$, or through a subsequence? Which smeared fields and normalization are used? On which topology of distributions is tightness proved?

3. Euclidean-to-relativistic reconstruction. Do the limiting Schwinger functions satisfy symmetry, regularity, reflection positivity, and the growth hypotheses needed by the chosen Osterwalder–Schrader theorem? The corrected reconstruction argument is not licensed by Euclidean invariance or pointwise positivity alone; see Osterwalder and Schrader 1975, § IV.2, pp. 288–304.

4. Nontriviality. Is the limiting field non-Gaussian, or does every truncated correlation of order greater than two vanish? Existence of a Gaussian limit is a genuine construction result, but it is not existence of interacting $\phi^4_4$.

This separation changes how the modern four-dimensional result is stated. For nearest-neighbor ferromagnetic Ising-type and Griffiths–Simon $\phi^4$ systems, the scaled fields are normalized by their variance and tested against compactly supported continuous functions. The limit is taken as $L\to\infty$ after the volume-to-observation-scale ratio tends to infinity, with subsequences and parameter variation allowed as specified in the definition. Under the stated susceptibility condition, every reachable four-dimensional scaling limit is a generalized Gaussian process Aizenman and Duminil-Copin 2021, Definition 1.1 and Theorem 1.2, pp. 166–167. Thus the theorem establishes Gaussianity of any limit in its class; it neither asserts convergence for every tuning path nor constructs a non-Gaussian continuum theory.

The physical setup and tuning language are developed on Bare Parameters, Tuning Conditions, and Continuum Targets. The theorem-first conclusion is narrower and sharper: a finite-cutoff calculation may support the first obligation and suggest a tuning, but it cannot by itself discharge tightness, reconstruction, or nontriviality.

Adversarial changes that alter the claim

Delete the topology. Suppose a source proves convergence of $n$-point distributions in $\mathcal S'((\mathbb R^4)^n)$ but the summary says only “the correlators converge.” It is then unclear whether convergence is pointwise away from diagonals, distributional, or strong in a Sobolev norm. These modes do not imply one another. The strongest surviving statement is the one with the source’s exact topology restored.

Interchange the limits. A proof of

$$\lim_{a\downarrow0}\left(\lim_{L\to\infty}S_{n,a,L}\right)=S_n$$

does not prove the reversed or joint limit. The first failed obligation is a uniform estimate controlling finite-volume errors as $a\downarrow0$ or cutoff errors as $L\to\infty$. Until such a bound is supplied, the altered statement is unsupported.

Strengthen existence to uniqueness. Tightness plus identification of one convergent subsequence proves neither convergence of the full family nor independence of boundary conditions. One needs uniqueness of all cluster points or another selection theorem.

Replace reconstruction hypotheses by a familiar example. The free massive covariance is reflection positive. That fact does not prove reflection positivity for an interacting or higher-derivative covariance. The property is an inequality for a whole positive-time test-function algebra, not a visual resemblance between two propagators.

Independent consistency checks

A record should be checked from more than one direction.

• Type check. Every map has a declared source and target. A distribution is never evaluated at a point unless a restriction theorem permits it; an unbounded operator is never multiplied without a common domain.
• Dimension and symmetry check. Both sides of each asserted relation transform in the same representation and have compatible physical dimensions.
• Limit check. A solvable Gaussian or finite-volume case reproduces the proposed normalization and order of limits.
• Negation check. Negate one hypothesis and exhibit either a counterexample or the exact proof step that stops. This is often more informative than repeating the successful derivation.
• Source check. The displayed conclusion can be located verbatim in the theorem, while any explanatory reformulation is visibly weaker, never stronger.

Common pitfalls

Treating a calculation as an existence proof. A long perturbative or numerical calculation can be correct within a regulator and still leave the limiting object unconstructed. Record the finite object and its error statement first, then list the additional compactness, renormalization, and reconstruction obligations.

Using “unique” without an equivalence relation. In QFT, uniqueness may mean equality of Schwinger functions, unitary equivalence of representations, natural isomorphism of functors, or independence of a renormalization choice up to local counterterms. State which one.

Reporting a theorem without its nonconverse. If $A\Rightarrow B$ is proved, the record should not let readers infer $B\Rightarrow A$. Give a counterexample when one is known, or say that the converse is unproved.

Exercises

1. Diagnose a limit statement. A sequence of probability measures $\mu_n$ has characteristic functionals $Z_n(f)$ satisfying $Z_n(f)\to Z(f)$ for every $f$ in a dense countable subset of $\mathcal S(\mathbb R^d)$. What is missing before one may claim $\mu_n\Rightarrow\mu$ on $\mathcal S'(\mathbb R^d)$?

Solution

One needs continuity and positive-definiteness of the limiting functional on the full test-function space to obtain a candidate measure, plus tightness of the family in a specified topology and an argument extending convergence from the dense subset. Without tightness, mass can escape in distribution space; without continuity, the pointwise limit need not be the characteristic functional of a Radon probability measure. The precise theorem used—typically a Minlos- and Lévy-type result on a nuclear space—must be stated with its topology.

2. Separate existence from interaction. Suppose all truncated Schwinger functions $S_n^T$ of a limiting scalar field vanish for $n>2$, while $S_2$ is nonzero. Which obligations above have been answered?

Solution

The nonzero two-point function shows that the limit is not the zero field, and vanishing higher truncated functions identifies a generalized Gaussian field. If the limiting family and OS hypotheses have actually been proved, existence and reconstruction may be answered. Nontriviality in the constructive-QFT sense of a non-Gaussian interacting field fails: Wick’s rule fixes all higher correlations from $S_2$.

3. Quantifier reversal. Construct a numerical array for which $\lim_{a\downarrow0}\lim_{L\to\infty}X_{a,L}$ exists but $\lim_{L\to\infty}\lim_{a\downarrow0}X_{a,L}$ is different.

Solution

Take $X_{a,L}=aL/(1+aL)$. For fixed $a>0$, the $L\to\infty$ limit is $1$, so the first iterated limit is $1$. For fixed $L$, the $a\downarrow0$ limit is $0$, so the reversed iterated limit is $0$. The example is elementary, but it proves that changing limit order is a mathematical change, not a stylistic rewrite.

References

• Aizenman, Michael, and Hugo Duminil-Copin. “Marginal Triviality of the Scaling Limits of Critical 4D Ising and $\phi^4_4$ Models.” Annals of Mathematics 194, no. 1 (2021): 163–235. DOI. Open PDF.
• Osterwalder, Konrad, and Robert Schrader. “Axioms for Euclidean Green’s Functions II.” Communications in Mathematical Physics 42 (1975): 281–305. DOI. Open PDF.