Clustering, Vacuum Assumptions, and Long-Range Correlations

Clustering is the statement that the connected part of a vacuum correlation disappears when two bounded local experiments are translated arbitrarily far apart in a spacelike direction. The limit depends on the chosen translation-invariant state and on the spectral channel seen by the operators: a mass gap can strengthen the decay to an exponential estimate, while gapless theories may still cluster algebraically. Degenerate vacua do not by themselves spoil clustering, but a non-extremal mixture of phases can retain a nonzero connected remainder at every distance.

Required background. Connected Correlators and Cumulants supplies the subtraction of disconnected one-point products and the distinction between full and connected functions. Multiparticle States, Statistics, and Fock Organization supplies the spectral-channel interpretation of one-particle shells and multiparticle continua.

Connected vacuum correlations at spacelike infinity

Let $\omega$ be a translation-invariant vacuum state, represented by a normalized vector $|\Omega\rangle$, and let

$$\alpha_a(B)=U(a)B U(a)^{-1}$$

translate a local observable $B$. For bounded local observables—or for controlled smeared fields on a common invariant domain—the connected two-point function is

$$C_{AB}(a) = \omega\!\left(A\,\alpha_a(B)\right) -\omega(A)\omega(B).$$

Choose a fixed spacelike vector $e$, so $e^2<0$, and translate by $a=\lambda e$. Once the translated localization regions are spacelike separated, the cluster property is

$$\lim_{\lambda\to+\infty}C_{AB}(\lambda e)=0.$$

This is an asymptotic factorization statement,

$$\omega\!\left(A\,\alpha_{\lambda e}(B)\right) \longrightarrow \omega(A)\omega(B),$$

not a claim that the correlation vanishes at a finite separation. It is also stronger than merely calling a correlator “connected”: taking the logarithm of a generating functional defines connected functions, while clustering asks how one of those functions behaves in a particular large-separation limit.

The vacuum sector controls the constant remainder

Center the observables,

$$A_0=A-\omega(A)\mathbf 1, \qquad B_0=B-\omega(B)\mathbf 1.$$

Because $U(a)|\Omega\rangle=|\Omega\rangle$,

$$C_{AB}(a) = \langle A_0^\dagger\Omega|U(a)B_0\Omega\rangle.$$

Let $P_0$ project onto the translation-invariant subspace. If the vacuum vector is the only invariant vector in this representation, then

$$P_0=|\Omega\rangle\langle\Omega|, \qquad P_0B_0|\Omega\rangle=0,$$

and therefore

$$C_{AB}(a) = \langle A_0^\dagger\Omega| (1-P_0)U(a)B_0\Omega\rangle.$$

The disconnected subtraction has removed the vacuum atom at zero four-momentum. Decay is then governed by the remaining translation spectrum and by the regularity supplied by locality and smearing. Vacuum uniqueness alone removes the constant vacuum contribution, but it is not by itself a pointwise decay theorem: singular continuous spectral weight or insufficient localization control still has to be excluded by the hypotheses of the chosen result.

If the invariant subspace contains additional vacuum directions, centering with respect to one state need not remove their contribution. That is the spectral reason vacuum purity or extremality matters in cluster statements.

A mass gap changes the available decay estimates

A representative relativistic cluster theorem assumes a unique invariant vacuum, local commutativity, the spectrum condition, controlled localization or smearing, and nonvacuum energy–momentum spectrum separated from the origin by an invariant mass $M>0$. In four spacetime dimensions, the Araki–Hepp–Ruelle estimate for equal-time spacelike separation $R$ has the characteristic form

$$|C_{AB}(R)| \leq c_{AB}\,R^{-3/2}e^{-MR},$$

with constants and finite localization offsets determined by the theorem’s setup. The result is a bound, not a universal asymptotic equality. Its hypotheses and theorem statement appear in Araki, Hepp, and Ruelle 1962, pp. 164–167; a later local-observable formulation sharpens the exponential cluster estimate in Fredenhagen 1985, pp. 461–463.

The lightest state in the theory need not control every operator pair. The leading exponent comes from the lightest spectral support to which both $A$ and $B$ couple. A symmetry may forbid a one-particle matrix element, or the first accessible contribution may be a multiparticle threshold. Thus a theory-wide gap gives a possible lower scale for decay, whereas the operator channel fixes the actual leading behavior.

Conversely, no mass gap does not imply failure of clustering. If the spectral measure reaches zero mass but has no invariant atom and has suitable infrared behavior, the connected correlation can still tend to zero, commonly as a power law. The exact power depends on dimension, field content, operator, and state.

Free scalar: exponential and algebraic regimes

Consider a free real scalar in four-dimensional Minkowski spacetime and an equal-time separation $x=(0,\mathbf r)$ with $r=|\mathbf r|>0$. The vacuum one-point function is zero, so the full and connected two-point functions agree. For $m>0$,

$$\begin{aligned} C_m(r) &= \langle\Omega|\phi(0,\mathbf r)\phi(0,\mathbf 0)|\Omega\rangle_c \\ &= \int\frac{\mathrm d^3\mathbf p}{(2\pi)^3\,2E_{\mathbf p}} e^{i\mathbf p\cdot\mathbf r} = \frac{m}{4\pi^2r}K_1(mr), \qquad E_{\mathbf p}=\sqrt{|\mathbf p|^2+m^2}. \end{aligned}$$

The Bessel form, translated to the site’s metric convention, follows from Weinberg 1995, § 5.2, p. 202. For $mr\gg1$,

$$C_m(r) = \frac{\sqrt m}{4\sqrt2\,\pi^{3/2}} \frac{e^{-mr}}{r^{3/2}} \left[ 1+\frac{3}{8mr} +O\!\left((mr)^{-2}\right) \right].$$

At fixed $r$, the massless limit instead gives

$$C_0(r)=\frac{1}{4\pi^2r^2}.$$

The two limits use the small-argument expansion in NIST DLMF, Eq. 10.30.2 and the large-argument expansion in NIST DLMF, Eq. 10.40.2. The massive field clusters exponentially, while the massless field clusters algebraically. Since $[\phi]=1$ in four dimensions, both $\sqrt m\,r^{-3/2}$ and $r^{-2}$ have the required mass dimension two.

The limits are not uniform: taking $r\to\infty$ at fixed $m>0$ probes the exponential regime, whereas setting $m=0$ first produces a power law. A statement about “long distance” must therefore declare which parameters and regulators are held fixed.

The operator channel can be seen even in the same free theory. For the vacuum Wick square,

$$\langle :\!\phi^2\!:(x):\!\phi^2\!:(0)\rangle_c =2C_m(r)^2.$$

Its massive leading behavior is

$$\frac{m}{16\pi^3}\frac{e^{-2mr}}{r^3},$$

while at $m=0$ it is $1/(8\pi^4r^4)$. The elementary field accesses the one-particle shell at $m$; the Wick square first accesses a two-particle channel at $2m$. A mass gap therefore does not make every correlator decay with the same exponent.

Degenerate vacua and long-range order

Suppose an order parameter $\Phi$ has two translation-invariant pure phases $\omega_+$ and $\omega_-$ with

$$\omega_\pm(\Phi)=\pm v,$$

and suppose each selected phase clusters. Then

$$\omega_\pm\!\left(\Phi(x)\Phi(y)\right) \longrightarrow v^2,$$

but its connected remainder still vanishes:

$$C_\pm(x-y) \longrightarrow v^2-(\pm v)^2=0.$$

The nonzero limit of the full two-point function is only the disconnected one-point product. Degeneracy has not destroyed clustering inside a selected extremal phase.

Now form the symmetric mixture

$$\omega_{\mathrm{mix}} =\tfrac12(\omega_++\omega_-).$$

It has $\omega_{\mathrm{mix}}(\Phi)=0$, while

$$\omega_{\mathrm{mix}}\!\left(\Phi(x)\Phi(y)\right) \longrightarrow v^2, \qquad C_{\mathrm{mix}}(x-y)\longrightarrow v^2.$$

The mixture retains information about a common phase choice over arbitrarily large distances. This is failure of clustering caused by non-extremality, not by the mere existence of more than one vacuum. The infinite-volume and phase-selection limits must be controlled before the separation limit; Weinberg develops this distinction for spontaneously broken vacua in Weinberg 1996, § 19.1, pp. 163–167.

Massless excitations introduce a different possibility. They can slow the decay to an algebraic law without leaving a constant remainder. Goldstone modes, gauge constraints, boundaries, lower-dimensional infrared behavior, thermal states, and finite-density states each require their own assumptions; the vacuum theorem above should not be transferred to them unchanged.

What clustering does not imply

Microcausality is not clustering. Spacelike local observables may commute while their connected correlation is nonzero. Microcausality constrains order; clustering constrains an asymptotic magnitude.

Clustering is not statistical independence. The limit does not provide a tensor-product factorization, a product state at finite distance, or absence of entanglement.

A gap is not an isolated particle pole. A spectral gap can begin at a continuum threshold. Establishing a stable one-particle shell requires additional spectral information.

Connected vacuum correlations are not 1PI functions or connected scattering amplitudes. All three use subtraction or connectedness language, but they organize different objects. Weinberg’s scattering cluster principle concerns far-separated multiparticle processes, not the limit defined on this page Weinberg 1995, § 4.3, pp. 177–181.

Point fields are distributional. The displayed free-field functions are evaluated away from coincidence. General structural statements use separated smeared fields or bounded local observables.

Check your understanding

1. Did the massless scalar fail to cluster? No. Since $C_0(r)=1/(4\pi^2r^2)\to0$, it clusters algebraically. What failed was the exponential estimate based on a positive mass gap.

2. Why does the symmetric vacuum mixture fail? Its one-point function vanishes by cancellation between phases, but its two-point function remembers that both distant insertions sample the same phase. The subtraction therefore leaves $v^2$.

3. Which mass sets the Wick-square exponent? The vacuum Wick square cannot create the free theory’s one-particle state from the vacuum. Its first accessible channel contains two particles, so $2C_m(r)^2$ begins with $e^{-2mr}$ rather than $e^{-mr}$.

Where to continue

• Develop the spectral measure: The Källén–Lehmann Representation relates isolated shells, continuum thresholds, and positive spectral weight to two-point functions.
• State theorem-level hypotheses before using a converse: the chapter’s structural hypothesis matrix distinguishes clustering, vacuum uniqueness, and mass-gap assumptions; Clustering, Vacuum Uniqueness, and the Mass Gap develops proof-level weak and strong cluster theorems, transfer-Hamiltonian results, and converse questions.
• Keep scattering connectedness separate: Scattering develops multiparticle amplitudes, diagrammatic connectedness, and the cluster properties appropriate to asymptotic processes.
• Return to the structural map: Structural Principles and Axiom Maps compares clustering with covariance, positivity, locality, and the theorem packages that use them.

References

• Araki, Huzihiro, Klaus Hepp, and David Ruelle. “On the Asymptotic Behaviour of Wightman Functions in Space-Like Directions.” Helvetica Physica Acta 35, no. 3 (1962): 164–174. Stable article record.
• Fredenhagen, Klaus. “A Remark on the Cluster Theorem.” Communications in Mathematical Physics 97 (1985): 461–463. DOI.
• National Institute of Standards and Technology. Digital Library of Mathematical Functions, Chapter 10, “Bessel Functions,” Eqs. 10.30.2 and 10.40.2. Small-argument formula; large-argument formula.
• Weinberg, Steven. The Quantum Theory of Fields. Volume I: Foundations. Cambridge University Press, 1995. DOI.
• Weinberg, Steven. The Quantum Theory of Fields. Volume II: Modern Applications. Cambridge University Press, 1996. DOI.