Non-Perturbative Construction and Non-Triviality of 4D Euclidean Phi4 Quantum Field Theory

Non-Perturbative Construction and Non-Triviality of 4D Euclidean Phi4 Quantum Field Theory

Definitions and mathematical statement

A Euclidean scalar field is a probability law on generalized functions on ${\mathbb{R}}^4$, or equivalently a compatible collection of Schwinger correlation distributions. The formal quartic action is \[ S(\phi)=\int_{{\mathbb{R}}^4}\left(\tfrac12|\nabla\phi|^2+\tfrac12m^2\phi^2+\lambda\phi^4\right)dx. \] This expression by itself does not define the continuum measure. A construction must specify regularization, renormalization, and a limit satisfying the Osterwalder--Schrader conditions, including Euclidean invariance, reflection positivity, symmetry, and regularity. Non-Gaussian means that the correlations are not all determined by the two-point function through Wick's rule. Source-fit warning: the cited Aizenman--Duminil-Copin theorem proves Gaussian limits for its four-dimensional ferromagnetic lattice Ising/$\phi^4$ class. The requested non-Gaussian construction cannot use that class of limits; the catalogue does not specify an alternative class. The record is retained, not certified as a well-posed currently open construction problem.

Formulation and concept references: [S1], [E1].

Short English statement

Can the proposed four-dimensional quartic scalar theory be constructed as a genuinely non-Gaussian continuum field, outside the class already ruled out by the cited triviality theorem?

Sources

• S1 M. Aizenman, H. Duminil-Copin, Ann. of Math. 194$1$:163-235, 2021. https://doi.org/10.4007/annals.2021.194.1.3. Catalogue formulation source.
• S2 Marginal triviality of the scaling limits of critical 4D Ising and phi4 models. https://annals.math.princeton.edu/2021/194-1/p03. Catalogue research/status reference; not a proof endorsement.
• S3 The non triviality of a Phi4 model, III, the Osterwalder-Schrader Positivity. https://arxiv.org/abs/2101.11980. Catalogue research/status reference; not a proof endorsement.
• S4 Triviality in a Non-Perturbative Second-Order Mean-Field Theory for phi4. https://arxiv.org/abs/2608.05998. Catalogue research/status reference; not a proof endorsement.
• E1 Michael Aizenman and Hugo Duminil-Copin, Marginal triviality of the scaling limits of critical 4D Ising and phi4 models, Annals of Mathematics 194 $2021$, 163--235; arXiv:1912.07973. https://arxiv.org/abs/1912.07973. Added primary source: Source-fit warning and proved triviality for the paper's model class. Consulted 24 September 2026.
• E2 Romain Panis, Triviality of the scaling limits of critical Ising and $\phi^4$ models with effective dimension at least four, arXiv:2309.05797; Annals of Probability, DOI 10.1214/25-AOP1782. Assumptions $A1$--$A6$, Theorems 1.12--1.13 and Corollary 1.14. https://arxiv.org/abs/2309.05797. Added research source: Quantitative triviality for the specified long-range model class; not a no-go theorem for arbitrary continuum constructions. Consulted 24 September 2026.
• E3 Francesca Arici, Daniel Becker, Chris Ripken, Frank Saueressig and Walter D. van Suijlekom, Reflection positivity in higher derivative scalar theories, arXiv:1712.04308 $2017$, Theorem 3.7; DOI 10.1063/1.5027231. https://arxiv.org/abs/1712.04308. Added research source: Rational-covariance reflection-positivity criterion. The negative reflection form used here is also derived explicitly in the attempt. Consulted 24 September 2026.
• E4 Majdouline Borji, Triviality in a Non-Perturbative Second-Order Mean-Field Theory for $\phi^4$, arXiv:2608.05998 (August 2026), concluding discussion of the ultraviolet remainder. https://arxiv.org/abs/2608.05998. Added research source: Mean-field result and its stated limitation; no assertion of full-theory triviality is imported. Consulted 24 September 2026.
• E5 Marietta Manolessou, The "non triviality" of a $\Phi^4_4$ model, III, the "Osterwalder--Schrader Positivity", arXiv:2101.11980 $2021$, Theorem 3.2. https://arxiv.org/abs/2101.11980. Added research source: Claimed positivity result. The preceding hierarchy construction and the full construction claim were not independently verified in this notebook; not a proof endorsement. Consulted 24 September 2026.

Non-Perturbative Construction and Non-Triviality of 4D Euclidean Phi4 Quantum Field Theory

Definitions and mathematical statement

A Euclidean scalar field is a probability law on generalized functions on ${\mathbb{R}}^4$, or equivalently a compatible collection of Schwinger correlation distributions. The formal quartic action is \[ S(\phi)=\int_{{\mathbb{R}}^4}\left(\tfrac12|\nabla\phi|^2+\tfrac12m^2\phi^2+\lambda\phi^4\right)dx. \] This expression by itself does not define the continuum measure. A construction must specify regularization, renormalization, and a limit satisfying the Osterwalder--Schrader conditions, including Euclidean invariance, reflection positivity, symmetry, and regularity. Non-Gaussian means that the correlations are not all determined by the two-point function through Wick's rule. Source-fit warning: the cited Aizenman--Duminil-Copin theorem proves Gaussian limits for its four-dimensional ferromagnetic lattice Ising/$\phi^4$ class. The requested non-Gaussian construction cannot use that class of limits; the catalogue does not specify an alternative class. The record is retained, not certified as a well-posed currently open construction problem.

Formulation and concept references: [S1], [E1].

Short English statement

Can the proposed four-dimensional quartic scalar theory be constructed as a genuinely non-Gaussian continuum field, outside the class already ruled out by the cited triviality theorem?

Research attempt

Current frontier.

The first obstruction is a theorem, not merely perturbative evidence. Aizenman--Duminil-Copin's Theorem 1.2 applies to the scaling-limit construction in their Section 1.2, with its Griffiths--Simon single-site measures, ferromagnetic nearest-neighbour interaction, normalization and critical/near-critical hypotheses. Within that construction the limiting field is Gaussian. This does not quantify over every possible meaning of a continuum quartic theory. [E1] Panis extends quantitative Gaussianity to specified long-range models with effective dimension at least four; his assumptions $A1$--$A6$ and the four-dimensional conclusion in Corollary 1.14 matter. Merely replacing nearest-neighbour interactions by a longer-range ferromagnetic kernel is therefore not an unrestricted escape route. [E2]

Two catalogue references require a different epistemic status. Manolessou's Theorem 3.2 claims reflection positivity for a previously constructed hierarchy; that earlier construction and the full continuum package have not been independently verified here. It is not used as an established solution. [S3], [E5] Borji's August 2026 analysis concerns a closed second-order mean-field hierarchy; its concluding discussion explicitly leaves the ultraviolet remainder outside the result. It supplies neither a general construction nor a universal no-go theorem. [S4], [E4]

Attack route.

Three routes were compared. Staying in the standard lattice class would require contradicting the cited Gaussianity theorem, so that route is excluded under its hypotheses. A positive covariance with stronger ultraviolet decay might make a quartic measure easier to construct, but it must retain reflection positivity. A cutoff-dependent perturbation of a known Gaussian-limit measure might instead retain positivity while changing the limit; that requires a perturbation that remains visible on continuum observables. The calculations below investigate the latter two routes and then test a tempting non-Gaussian construction based on mixing free fields. They are deductions made for this notebook, not claims of literature novelty.

Attempt: a spectral obstruction to ultraviolet softening.

Lemma 51.1 (positive spectral mass cannot disappear at high momentum). Suppose a scalar covariance multiplier has the unsubtracted representation \begin{equation}\label{wsref-0-72-61-3a-33-30-30-30-36-36-39-37-3a-73-74-69-65-6c-74-6a-65-73} c(s)=\int_{[0,\infty)}\frac{\rho(\,\mathrm d t)}{s+t},\qquad s>0, \qquad 0\ne\rho\ge0, \end{equation} and the integral is finite for each $s>0$. Then \[ \lim_{s\to\infty}s\,c(s)=\rho([0,\infty))\in(0,\infty]. \] Indeed, $s/(s+t)$ increases to one at every finite $t$, so the monotone convergence theorem applies, including when the total mass is infinite. In particular, $c(s)=o(s^{-1})$ is impossible under (ra:30006697:stieltjes). The representation is an explicit hypothesis: this argument does not assume that every distributional covariance admits an unsubtracted representation of this form.

Consider the concrete higher-derivative proposal \[ c_\varepsilon(s)=\frac{1}{s+m^2+\varepsilon s^2},\qquad m>0,\quad 0<4\varepsilon m^2<1. \] Put \[ a=\frac{1-\sqrt{1-4\varepsilon m^2}}{2\varepsilon},\qquad b=\frac{1+\sqrt{1-4\varepsilon m^2}}{2\varepsilon}. \] Then $0<a<b$ and the exact partial-fraction identity is \begin{equation}\label{wsref-0-72-61-3a-33-30-30-30-36-36-39-37-3a-67-68-6f-73-74} c_\varepsilon(s)=\frac{1}{\varepsilon(b-a)} \left(\frac{1}{s+a}-\frac{1}{s+b}\right). \end{equation} The negative residue is consistent with the general rational-propagator criterion of Arici et al., Theorem 3.7. Here it is useful to exhibit the negative reflection form directly rather than merely invoke the criterion. [E3]

Reflect the first Euclidean coordinate, and Fourier transform in the remaining three. For fixed spatial momentum $k$, set $\omega_a=(|k|^2+a)^{1/2}$, $\omega_b=(|k|^2+b)^{1/2}$, and $\mathcal Lf(\omega)=\int_0^\infty e^{-\omega t}f(t)\,\mathrm d t$. With the usual inverse Fourier normalization, the reflected quadratic form for a temporal test function is \begin{equation}\label{wsref-0-72-61-3a-33-30-30-30-36-36-39-37-3a-72-65-66-6c-65-63-74-69-6f-6e} Q_k(f)=\frac{1}{\varepsilon(b-a)} \left(\frac{|\mathcal Lf(\omega_a)|^2}{2\omega_a} -\frac{|\mathcal Lf(\omega_b)|^2}{2\omega_b}\right). \end{equation} This follows by inverse transforming $(p_0^2+\omega^2)^{-1}$, whose temporal kernel is $e^{-\omega|t|}/(2\omega)$; reflection makes the argument $t+s$.

Choose a nonnegative nonzero $h\in C_c^\infty(0,\infty)$ and $0<t_1<t_2$. For a fixed $k_0$, take \[ f(t)=h(t-t_1)-e^{\omega_a(k_0)(t_2-t_1)}h(t-t_2). \] Its Laplace transform vanishes at $\omega_a(k_0)$ and is nonzero at $\omega_b(k_0)$, since these frequencies are different. Thus $Q_{k_0}(f)<0$. By continuity the inequality persists in a neighbourhood of $k_0$. Multiplying by a nonzero smooth spatial Fourier test function supported in that neighbourhood and integrating gives a negative reflection form for a Schwartz test function supported at positive time. Therefore this particular free regulator fails reflection positivity at every indicated finite $\varepsilon$, despite its positive Fourier multiplier on real momenta. No numerical approximation is involved.

This rejects a proposed positive-metric regulator, not the target theory. As $\varepsilon\downarrow0$, the heavy pole moves to infinity; reflection positivity might conceivably be restored in a limit. An interacting construction using such regulators would still have to prove tightness, renormalized convergence, limiting reflection positivity and a nonzero connected correlation. The finite-cutoff calculation establishes none of those limiting assertions.

Attempt: how large must a perturbation be to escape Gaussianity?

Lemma 51.2 (change-of-measure bound). Let $\mu_a$ be cutoff probability measures and let $\nu_a=r_a\mu_a$, where $r_a\ge0$ and $\int r_a\,\mathrm d\mu_a=1$. Fix $1<p<\infty$, $q=p/(p-1)$, and write $\delta_a=\|r_a-1\|_{L^p(\mu_a)}$. Let $Y_{1,a},\ldots,Y_{4,a}$ be real observables, centred under both measures, satisfying \[ \max_i\|Y_{i,a}\|_{L^{4q}(\mu_a)}\le M \] with $M$ independent of $a$. Their connected fourth moments satisfy \begin{equation}\label{wsref-0-72-61-3a-33-30-30-30-36-36-39-37-3a-63-75-6d-75-6c-61-6e-74-62-6f-75-6e-64} |\kappa_{4,\nu_a}(Y_{1,a},\ldots,Y_{4,a}) -\kappa_{4,\mu_a}(Y_{1,a},\ldots,Y_{4,a})| \le M^4(7\delta_a+3\delta_a^2). \end{equation}

To prove this, Hölder's inequality gives \[ |{\mathbb{E}}_{\nu_a}Y_1Y_2Y_3Y_4-{\mathbb{E}}_{\mu_a}Y_1Y_2Y_3Y_4| \le\delta_a M^4, \qquad |{\mathbb{E}}_{\nu_a}Y_iY_j-{\mathbb{E}}_{\mu_a}Y_iY_j| \le\delta_a M^2. \] The second estimate also uses monotonicity of $L^r$ norms on a probability space. Every reference covariance has absolute value at most $M^2$. Writing a perturbed covariance as $C_{ij}+\Delta_{ij}$, the difference of each product of two covariances is at most $M^4(2\delta_a+\delta_a^2)$. Subtract the three pairing products from the fourth moment. This proves (ra:30006697:cumulantbound). Centring is not silently inherited after tilting: for example, it is guaranteed when both measures are invariant under field sign reversal and all $Y_i$ are odd. Without such a condition additional mean terms must be included.

There is also a stronger qualitative observation that needs no moment bound. For any finite vector $Y_a$ of smeared observables and $t\in{\mathbb{R}}^j$, \begin{equation}\label{wsref-0-72-61-3a-33-30-30-30-36-36-39-37-3a-63-68-61-72-61-63-74-65-72-69-73-74-69-63} \left|{\mathbb{E}}_{\nu_a}e^{i t\cdot Y_a}-{\mathbb{E}}_{\mu_a}e^{i t\cdot Y_a}\right| \le\|r_a-1\|_{L^1(\mu_a)}\le\delta_a. \end{equation} Consequently, if $\delta_a\to0$, every existing Gaussian finite-dimensional limit under $\mu_a$ remains the same under $\nu_a$. This statement uses bounded characteristic functions, not an unjustified inference from weak convergence to convergence of fourth moments. Applied to the lattice limits of [E1], it says that asymptotically negligible density tilts cannot provide the requested non-Gaussian limit.

For a quantitative witness, suppose the reference fourth cumulant tends to zero and the absolute limiting perturbed cumulant is at least $\eta>0$. Under the uniform moment hypothesis, (ra:30006697:cumulantbound) forces \[ \liminf_{a\to0}\delta_a \ge \frac{\sqrt{49+12\eta/M^4}-7}{6}>0. \] This is a necessary condition, not a sufficient construction criterion. A small bare coefficient in $r_a=Z_a^{-1}e^{-V_a}$ does not imply $\delta_a\to0$: the growing volume, ultraviolet fluctuations and the normalizing partition function all intervene. Conversely, the lemma does not rule out a perturbation large in microscopic total variation but small on a particular finite set of smeared observables.

Stress test: a non-Gaussian field that does not solve the problem.

Let $\Psi$ be a centred massive free Euclidean field and let $V$ be an independent positive, bounded, nonconstant random variable with ${\mathbb{E}} V=1$. The field $\Phi=\sqrt V\,\Psi$ is non-Gaussian. Conditional Wick expansion gives, for $C_{ij}={\mathbb{E}}\Psi(f_i)\Psi(f_j)$, \[ \kappa_4(\Phi(f_1),\ldots,\Phi(f_4)) =\operatorname{Var}(V) (C_{12}C_{34}+C_{13}C_{24}+C_{14}C_{23}). \] Euclidean invariance and reflection positivity survive the mixture: the reflected expectation of $\overline{F(\theta\Phi)}F(\Phi)$ is an average of nonnegative free-field reflected expectations. This elementary construction therefore defeats the false claim that reflection positivity alone forces Gaussianity.

It also exposes why those checks are insufficient. If $g_R$ is a large translate of $g$, the massive free cross-covariance tends to zero, whereas \[ \operatorname{Cov}\left(\Phi(f)^2,\Phi(g_R)^2\right) \longrightarrow \operatorname{Var}(V)\,C(f,f)\,C(g,g)>0. \] Thus the mixture has a persistent global random scale and fails the clustering requirement in the usual vacuum formulation. Independently of which clustering convention is intended, no local quartic interaction has been exhibited. The mixture is not a candidate answer to Section 30006697.

The two substantive failed routes are now precise. Stronger ultraviolet falloff conflicts with a positive unsubtracted spectral measure, and the explicit higher-derivative covariance has a negative reflected direction. Small measure tilts preserve a Gaussian scaling limit under the stated hypotheses. Neither argument covers all local non-Gaussian continuum constructions. In particular, absence of a fourth cumulant would not by itself prove an arbitrary law Gaussian; the full characteristic-function argument above is what establishes Gaussianity for negligible tilts.

Outcome.

Outcome: Exploratory approach with unresolved gap.

The obtained results are the explicit negative-reflection test, the quantitative fourth-cumulant stability bound, and the diagnosis of the Gaussian-mixture false construction. They constrain proposed escapes from the established lattice triviality theorem but do not produce a quartic continuum field or rule out every such field. A successful continuation must specify an alternative regularized model, retain an order-one non-Gaussian effect on continuum observables, and prove all required continuum axioms. The unspecified alternative model in the original entry remains a genuine formulation limitation, not a licence to substitute an arbitrary non-Gaussian random field.

Sources

• S1 M. Aizenman, H. Duminil-Copin, Ann. of Math. 194$1$:163-235, 2021. https://doi.org/10.4007/annals.2021.194.1.3. Catalogue formulation source.
• S2 Marginal triviality of the scaling limits of critical 4D Ising and phi4 models. https://annals.math.princeton.edu/2021/194-1/p03. Catalogue research/status reference; not a proof endorsement.
• S3 The non triviality of a Phi4 model, III, the Osterwalder-Schrader Positivity. https://arxiv.org/abs/2101.11980. Catalogue research/status reference; not a proof endorsement.
• S4 Triviality in a Non-Perturbative Second-Order Mean-Field Theory for phi4. https://arxiv.org/abs/2608.05998. Catalogue research/status reference; not a proof endorsement.
• E1 Michael Aizenman and Hugo Duminil-Copin, Marginal triviality of the scaling limits of critical 4D Ising and phi4 models, Annals of Mathematics 194 $2021$, 163--235; arXiv:1912.07973. https://arxiv.org/abs/1912.07973. Added primary source: Source-fit warning and proved triviality for the paper's model class. Consulted 24 September 2026.
• E2 Romain Panis, Triviality of the scaling limits of critical Ising and $\phi^4$ models with effective dimension at least four, arXiv:2309.05797; Annals of Probability, DOI 10.1214/25-AOP1782. Assumptions $A1$--$A6$, Theorems 1.12--1.13 and Corollary 1.14. https://arxiv.org/abs/2309.05797. Added research source: Quantitative triviality for the specified long-range model class; not a no-go theorem for arbitrary continuum constructions. Consulted 24 September 2026.
• E3 Francesca Arici, Daniel Becker, Chris Ripken, Frank Saueressig and Walter D. van Suijlekom, Reflection positivity in higher derivative scalar theories, arXiv:1712.04308 $2017$, Theorem 3.7; DOI 10.1063/1.5027231. https://arxiv.org/abs/1712.04308. Added research source: Rational-covariance reflection-positivity criterion. The negative reflection form used here is also derived explicitly in the attempt. Consulted 24 September 2026.
• E4 Majdouline Borji, Triviality in a Non-Perturbative Second-Order Mean-Field Theory for $\phi^4$, arXiv:2608.05998 (August 2026), concluding discussion of the ultraviolet remainder. https://arxiv.org/abs/2608.05998. Added research source: Mean-field result and its stated limitation; no assertion of full-theory triviality is imported. Consulted 24 September 2026.
• E5 Marietta Manolessou, The "non triviality" of a $\Phi^4_4$ model, III, the "Osterwalder--Schrader Positivity", arXiv:2101.11980 $2021$, Theorem 3.2. https://arxiv.org/abs/2101.11980. Added research source: Claimed positivity result. The preceding hierarchy construction and the full construction claim were not independently verified in this notebook; not a proof endorsement. Consulted 24 September 2026.