Continuum Limits and Universality Problems

A continuum limit is a convergence theorem for a specified family of regulated objects. Universality is stronger: two or more microscopic families, after their own tuning and field renormalizations, lead to equivalent limiting objects and matching identified observables. Agreement of critical exponents, perturbative beta functions, or a few extrapolated correlators can support universality, but it does not prove convergence, uniqueness, or equivalence of complete quantum field theories.

Required background. The constructive program and cutoff removal supplies the limit sequence. Stable manifolds and relevant–marginal control supplies tuned trajectories. Lines of constant physics supply the practical extrapolation, and existence, uniqueness, and equivalence fix the conclusion. Helpful background. Euclidean measures and relativistic models and regulators and continuum limits give the reconstruction and physical context.

A continuum limit with all quantifiers shown

Let $\mu_{a,L,\mathbf g}$ be regulated probability measures, with lattice spacing $a$, volume $L$, and bare parameters $\mathbf g$. Choose a tuning curve $\mathbf g=\mathbf g(a)$ and a field renormalization $Z_a$. One possible theorem target is weak convergence

$$\mu_{a,L(a),\mathbf g(a)}\circ (Z_a^{-1/2}\phi_a)^{-1} \Longrightarrow \mu \quad\text{in }\mathcal S'(\mathbb R^d).$$

This notation contains decisions that prose must unpack. Does $L(a)\to\infty$ before, after, or jointly with $a\to0$? Is the topology weak convergence of probability laws on tempered distributions, convergence in a weighted Besov space, convergence of every smeared moment, or convergence of a net of algebras? Are composite operators renormalized and retained? Does the limiting law depend on the subsequence?

Tightness provides subsequential limits, not uniqueness. Convergence of characteristic functionals can identify a law if continuity and positivity hold. Moment convergence identifies the law only with a determinacy or uniform-integrability argument. A QFT conclusion adds Euclidean covariance, reflection positivity, regularity, clustering, and OS reconstruction. These are separate proof steps.

What universality would prove

Let $A$ and $B$ be two discretizations. A strong universality statement supplies tuning maps $\mathbf g_A(a),\mathbf g_B(b)$, field and composite-operator renormalizations, limits $\mu_A,\mu_B$, and an isomorphism or unitary equivalence

$$U:\mathcal Q_A\xrightarrow{\simeq}\mathcal Q_B$$

that intertwines a declared generating set of observables. Matching a susceptibility exponent tests one coordinate of this statement. Matching finitely many low-point correlations still need not determine the complete measure or observable algebra.

The theorem class must also say which perturbations are irrelevant. If two lattice actions differ by a symmetry-breaking relevant operator that was not tuned away, they need not share a limit. “Same fixed point” is a conclusion of the controlled RG flow, not a label attached before the flow is proved.

Rigorous examples and their boundaries

In two-dimensional critical Ising theory, Camia, Garban, and Newman prove existence and uniqueness of the magnetization-field scaling limit as a random distribution, with conformal covariance, for the stated lattice model and normalization 2015, Theorems 1.1–1.2, pp. 528–571. This constructs a specific scaling field. It does not by itself identify every lattice observable or every model in the Ising universality class.

In four dimensions, Aizenman and Duminil-Copin prove that the normalized scaling limits in their nearest-neighbor ferromagnetic Ising-type and lattice $\lambda\phi^4$ classes are Gaussian 2021, Theorem 1.2, pp. 163–177. This is a universality result toward the Gaussian field within the stated class; logarithmic normalizations and thermodynamic corrections remain nontrivial.

The $SU(2)$ Yang–Mills–Higgs theorem supplies a different controlled limit. Under a specific joint scaling of $a$, gauge coupling, and Higgs length, a projected gauge field converges to a massive Gaussian random one-form Chatterjee 2026, Theorems 3.1–3.2, pp. 10–14. The source explicitly leaves slower-coupling and non-Gaussian limits open. Changing the tuning curve is therefore a new problem, not a harmless regulator change.

First application: two three-dimensional scalar discretizations

Return to bare parameters, tuning conditions, and continuum targets. Suppose a nearest-neighbor and an improved lattice action are simulated near the same proposed three-dimensional scalar fixed point. A theorem-level comparison would require:

1. a critical surface and tuned relevant parameters for each action;
2. a common normalization for the scaling field;
3. tightness in one distribution topology;
4. uniqueness of each limiting law;
5. convergence of a separating collection of local and composite observables;
6. an identification map preserving those observables and the OS structure.

If only the critical exponent $\nu$ and one amplitude ratio agree, the strongest conclusion is quantitative universality evidence for those observables. It is not yet equality of limiting QFTs. If a rigorous RG theorem controls both actions in one stable neighborhood and proves their irrelevant coordinates contract to the same renormalized trajectory, it can supply the missing mechanism—but the norm, basin, and observable recovery must be stated.

Ordered limits and noncommutation

Volume and continuum limits need not commute. Taking $L\to\infty$ first can select a phase; keeping $L$ proportional to the correlation length produces finite-size scaling; taking $a\to0$ at fixed physical volume yields a continuum theory in a box. A massless limit after reconstruction may differ from a massless bare sequence. The theorem must display the order or prove commutation uniformly.

Failure test: matching exponents

Observe identical exponents over several lattice spacings and declare identical continuum QFTs. The test asks for tightness, full correlation convergence, uniqueness, and an observable-preserving map. Without them, a distinct theory could share the measured exponent or the apparent scaling window could cross over. The surviving conclusion is evidence restricted to the measured observables and range.

Exercises

Why does tightness plus convergence of the two-point function not establish a unique continuum measure?

Solution

Tightness gives subsequential laws, and a common two-point function fixes only their covariance. Distinct non-Gaussian measures can share that covariance. One needs a determining characteristic functional or full moment hierarchy with uniqueness conditions, together with proof that every subsequence has the same limit.

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 (2021): 163–235; corrigendum 199 (2024): 479. DOI.
• Camia, Federico, Christophe Garban, and Charles M. Newman. “Planar Ising Magnetization Field I. Uniqueness of the Critical Scaling Limit.” Annals of Probability 43 (2015): 528–571. DOI; Open PDF.
• Chatterjee, Sourav. “A Scaling Limit of $SU(2)$ Lattice Yang–Mills–Higgs Theory.” Probability and Mathematical Physics 7 (2026): 339–381. DOI; Open PDF.