Euclidean Growth, Regularity, and Temperedness Conditions

Regularity in Euclidean reconstruction has three levels that must not be conflated: each $S_n$ must act continuously on its test space; the continuity estimates must be controlled across the whole hierarchy; and the analytically continued functions must admit tempered Lorentzian boundary values. The corrected OS theorem uses a hierarchy-wide growth condition precisely because fixed-$n$ temperedness does not control the last two steps.

Required background. Euclidean random fields and Schwinger hierarchies supplies moments and characteristic functionals. Domains, signatures, supports, and regularity supplies the domain distinctions. Test functions, distributions, and support supplies distributional continuity. Tempered distributions and Fourier calculus supplies Schwartz seminorms and boundary values.

Helpful background. Locally convex, nuclear, and rigged Hilbert spaces explains why nuclear test spaces support measure extension and kernel theorems.

Fixed-order distributional control

For each $n$, a tempered Schwinger distribution is a continuous linear functional

$$S_n\in\mathcal S'\!\left((\mathbb R^d)^n\right).$$

Equivalently, there exist constants $C_n$ and a finite seminorm order $k_n$ such that

$$\lvert S_n(f)\rvert \leq C_n\max_{\lvert\alpha\rvert\leq k_n} \sup_x(1+\lvert x\rvert)^{k_n} \lvert\partial^\alpha f(x)\rvert.$$

This bound is meaningful only after specifying whether $f$ is a general Schwartz function on all configurations or belongs to a subspace vanishing at coincidence diagonals. Composite-field singularities can make that distinction consequential. Separate-variable continuity of an $n$-linear smeared functional is also not a substitute for the joint continuity needed to invoke the Schwartz kernel theorem, although nuclear Fréchet spaces provide useful equivalences under standard hypotheses.

Temperedness permits Fourier transformation and polynomial growth at infinity. It does not imply that $S_n$ is a function, that it can be restricted to a sharp time slice, or that its pointwise values exist away from every singular set. Those are additional regularity statements.

Uniform control across the hierarchy

The original OS distribution condition controlled each order but did not supply enough uniformity for the claimed converse theorem. OS II introduced alternatives E0′ and E0″. In the E0′ form, there are a fixed integer $s$ and constants $\sigma_n$ of factorial growth such that, schematically,

$$\lvert S_n(f)\rvert\leq\sigma_n\lVert f\rVert_{sn}, \qquad \sigma_n\leq A(n!)^B.$$

E0″ uses a stronger product-type distribution bound. The precise seminorms and admissible test spaces are given in Osterwalder and Schrader 1975, §IV.1 and Appendix, pp. 287–288, 303–305. These are sufficient conditions for that theorem; the page does not claim that every reconstructible theory must satisfy this particular formulation.

Why does the dependence on $n$ matter? Reconstruction first obtains analytic functions in restricted ordered domains and then enlarges those domains by repeated continuation. Both the number of continuation steps and the distributional order can grow with $n$. A bound that is harmless for each fixed $n$ may become useless if $k_n$ or $C_n$ grows without control. Factorial growth is compatible with the combinatorics of field moments while still allowing the boundary-value estimates to close.

Moment bounds and distribution bounds answer different questions. An estimate such as

$$\mathbb E[\lvert\Phi(f)\rvert^n]\leq n!\,K_f^n$$

controls one smeared random variable and can imply local analyticity of its characteristic function. It does not uniformly control $S_n(f_1\otimes\cdots\otimes f_n)$ as the supports, derivatives, and relative positions of all $f_j$ vary. Conversely, a distributional OS bound does not by itself construct a probability measure on $\mathcal S'$.

Regularity away from coincidences

Euclidean covariance, reflection positivity, and the other OS conditions imply strong analyticity away from coincident points in the reconstruction argument. One works first on ordered regions such as

$$\tau_1<\tau_2<\cdots<\tau_n,$$

where semigroup matrix elements are available. The functions extend to complex domains, while diagonals and their complex continuations remain potential singular loci. OS II proves real analyticity in the relevant Euclidean regions before moving toward Lorentzian boundary values Osterwalder and Schrader 1975, §V.1, pp. 291–293.

A typical boundary-value theorem has this structure: if an analytic function $F(x+iy)$ in a tube obeys polynomial bounds in $x$ and controlled inverse-power growth as $y$ approaches the cone boundary, then

$$\lim_{y\to0,\ y\in C}F(x+iy)$$

exists in $\mathcal S'$. The direction $y\in C$, the component of the tube, and the topology of the limit are part of the conclusion. Pointwise substitution at $y=0$ is generally invalid.

First QFT application: constructive P(φ)₂ bounds

In the weakly coupled massive $P(\phi)_2$ model, ultraviolet regularization, Wick ordering, volume limits, and cluster expansions are used to construct Schwinger functions rather than postulate them. The estimates have to do more than show that every smeared moment is finite: they control insertion number, derivatives or Schwartz seminorms, volume, and separation uniformly enough to pass to infinite volume and then to tempered Lorentzian distributions.

A particularly transparent modern estimate is available for the planar $P(\Phi)_2$ measure. For every accumulation point $\mu$ constructed there, one finds a ball $B$ about zero in a Schwartz seminorm such that, for the even degree $r$ of the polynomial interaction,

$$\int_{\mathcal S'}\exp\!\left(\Phi(f)^r\right)d\mu(\Phi)\leq2, \qquad f\in B.$$

Rigorous construction status and open problems tracks the dimensional and model-dependent boundary of this application.

Duch, Dybalski, and Jahandideh 2025, Theorem 1.1 and Remarks 1.3–1.4 prove that the displayed exponential integrability gives OS regularity, together with Euclidean invariance and reflection positivity. Their construction does not establish clustering in general, so it does not by itself give vacuum uniqueness.

A representative target bound has the form

$$\lvert S_n(f_1,\ldots,f_n)\rvert \leq A^n(n!)^B\prod_{j=1}^n\lVert f_j\rVert_s,$$

with $A,B,s$ independent of $n$ after the cutoffs are removed. Cluster estimates additionally control connected functions when supports separate. Such estimates verify the kind of hierarchy-wide regularity needed by OS reconstruction; they are not obtained merely by expanding $e^{-\int P(\phi)}$ formally.

For small coupling, the constructed $P(\phi)_2$ model was proved to satisfy the Wightman axioms and to possess a rigorously analyzed particle spectrum Glimm, Jaffe, and Spencer 1974, pp. 585–632. That is a model- and regime-specific constructive result. It does not establish continuum existence for four-dimensional polynomial scalar theory, nor does it make a formal perturbation series into a measure.

Adversarial growth test

Suppose every $S_n$ is tempered and its scalar moments grow only factorially, but the smallest usable distribution order grows as $k_n=n^2$. For example, one can arrange increasingly high derivatives of a fixed tempered kernel while normalizing selected scalar moments. Each order separately has a bound, yet there is no fixed $s$ for which $k_n\leq sn$. The family therefore fails E0′ even though the phrases “tempered at every order” and “factorial moment growth” are both true.

This failure does not prove that no alternative reconstruction is possible. It proves only that the hypotheses of the selected OS II theorem have not been verified. A valid argument must either establish the required uniform estimate or cite a different theorem with hypotheses the hierarchy actually satisfies.

Independent checks

• Test space: record whether kernels may meet coincidence diagonals and whether sharp-time restriction is used.
• Uniformity: identify every constant’s dependence on $n$, cutoff, volume, support separation, and seminorm order.
• Topology: state whether convergence is pointwise, weak distributional, strong distributional, or in a measure topology.
• Boundary direction: verify that the analytic estimate is uniform inside the cone used for the Lorentzian limit.
• Model scope: separate a finite-cutoff estimate from a cutoff-independent continuum estimate.

Exercise

Explain why $S_n=\partial^{n^2}\delta$ (in one chosen relative coordinate, with the other variables harmlessly smeared) is tempered for every $n$ but cannot satisfy a bound whose seminorm order grows at most linearly in $n$.

Solution

For fixed $n$, $\partial^{n^2}\delta$ acts continuously by

$$\lvert\partial^{n^2}\delta(f)\rvert =\lvert\partial^{n^2}f(0)\rvert,$$

so it is tempered of order $n^2$. A Schwartz seminorm controlling derivatives only through order $sn$ cannot bound the derivative of order $n^2$ uniformly once $n>s$. Test functions can be chosen with all derivatives through order $sn$ bounded while the $n^2$-th derivative at the origin grows. Thus fixed-order temperedness does not imply the E0′ linear-order control.

References

• Duch, Paweł, Wojciech Dybalski, and Azam Jahandideh. “Stochastic Quantization of Two-Dimensional $P(\Phi)$ Quantum Field Theory.” Annales Henri Poincaré 26 (2025): 1055–1086. doi:10.1007/s00023-024-01447-w. Open article.
• Glimm, James, Arthur Jaffe, and Thomas Spencer. “The Wightman Axioms and Particle Structure in the $\mathscr P(\phi)_2$ Quantum Field Model.” Annals of Mathematics 100 (1974): 585–632. doi:10.2307/1970959. Journal page.
• Osterwalder, Konrad, and Robert Schrader. “Axioms for Euclidean Green’s Functions II.” Communications in Mathematical Physics 42 (1975): 281–305. doi:10.1007/BF01608978. Open PDF.