Skip to Content
UniverseUniverse
  • Article
  • Open Access

23 October 2026

34 Pages

Infrared Divergences as Itinerant Vacua

1
RIKEN, Wako 351-0198, Saitama, Japan
2
Department of Physics, Ochanomizu University, Bunkyo, Tokyo 112-8610, Japan

Abstract

Infrared divergences (IRDs) are usually treated as pathologies to be canceled, regularized, or hidden in dressed asymptotic states. This paper develops a complementary and constructive viewpoint: an IRD is the signature of an itinerant vacuum—a quantum vacuum that wanders continuously through a family of inequivalent states as a classical order parameter evolves. Each value of the order parameter carries its own coherent vacuum, so moving the order parameter means traversing a succession of orthogonal vacua. The IRD is the field-theoretic cost of this wandering, and the 1 / f noise, gravitational memory, and non-Gaussian fluctuations that emerge from it are its observable classical remnants. The technical core is an exact separation in the real-time closed-time-path (CTP) effective action. For soft QED, we take the standard Faddeev–Kulish coherent dressing as the asymptotic, infrared-safe description; the new step is to promote its fixed in/out dressing profile α to a real-time history α [ J ( t ) ] along with the evolving current J ( t ) . The infrared-enhanced symmetric part of the corresponding reduced CTP influence functional is then converted, by a Hubbard–Stratonovich identity, into a classical stochastic source. This step is an algebraic identity of the generating functional and requires no additional coarse graining once the reduced CTP effective action for the collective variable has been defined: the retarded kernel encodes the memory of past vacuum transitions, while the noise kernel encodes the quantum uncertainty of the next one. We apply this construction to four parallel arenas—soft QED, scalar fields in de Sitter space, soft gravitons, and non-equilibrium phase transitions—and show that the same itinerant-vacuum mechanism underlies 1 / f current noise, the primordial power spectrum, gravitational memory, and order-parameter dynamics. A geometric formulation in terms of a Hilbert-space bundle over the vacuum manifold is outlined as an outlook.

1. Introduction

Infrared divergences are among the most persistent structures in quantum field theory. The conventional reactions—cancellation in inclusive quantities, dressed asymptotic states, finite resolution, or resummation of secular logarithms [1,2,3,4,5,6,7,8]—organize the infrared sector so that physical quantities are finite. We do not replace this standard infrared-safe scattering theory. In QED we use the Faddeev–Kulish (FK) coherent dressing as its asymptotic form and ask what the same soft cloud does during real-time evolution. A fixed in/out dressing α then becomes α [ J ( t ) ] . In the reduced CTP description, the infrared-enhanced symmetric part of this dynamical soft sector is relocated into a stochastic covariance by the Hubbard–Stratonovich identity. Once this is performed, the infrared enhancement is no longer merely a perturbative pathology; it becomes a possible origin of observable classical structures.
Three emergent phenomena deserve particular attention.
1.
Classical c-number random fields. The imaginary positive part of the CTP influence functional, when separated by a Hubbard–Stratonovich identity, produces a classical stochastic source. No prior measurement or decoherence process is assumed.
2.
1 / f fluctuations with a calculable amplitude. The soft-photon (or soft-graviton) phase-space integral gives a noise kernel N ( ω ) ∝ 1 / | ω | . This is the field-theoretic origin of 1 / f noise in current-carrying systems, whose amplitude is expressed in terms of source parameters: charge e, density n, velocity β , and scattering rate γ imp .
3.
Macroscopic order parameters. In a non-equilibrium phase transition or in inflationary cosmology, the unstable long-wavelength sector separates from microscopic fluctuations and becomes the stochastic noise that drives a classical order parameter. The “reason the universe chose a particular branch” is encoded in this separation, not in an externally imposed decoherence mechanism.
Historical context: Handel’s proposal. The idea that 1 / f current noise in conductors might be a fundamental QED infrared phenomenon was proposed by Handel in 1975 [9,10]. He argued that soft-photon radiative corrections to charged-particle scattering produce a 1 / f contribution to current fluctuations, with a universal coefficient set by the fine structure constant. This was a physically suggestive proposal: it identified a possible connection between the universality of low-frequency noise [11,12] and the universal soft structure of QED.
Handel’s theory was subsequently criticized on formal grounds by several authors [12,13,14]. The central objections are: (a) scattering states with different soft-photon content are orthogonal in the strict infrared limit, making interference between different soft sectors ill-defined and (b) a direct identification of infrared-divergent scattering cross sections with current fluctuations is not rigorously justified in an S-matrix framework.
The present paper revisits Handel’s central intuition—that soft electromagnetic radiation associated with charged-particle scattering is physically relevant to 1 / f noise—but formulates the mechanism differently and more rigorously. In our formulation, soft-sector orthogonality is not a contradiction; it is precisely the statement that tracing over the soft sector causes reduced-state decoherence. The route to 1 / f noise is not a direct cross-section argument, but a CTP-derived identity,
soft IRD ⟶ Im Γ IF ⟶ N IR ( ω ) ⟶ S J ( ω ) = A J | ω | .
The amplitude A J is determined by the Ward-identity-constrained kinematic parameters of the charged carriers, not by a universal constant. In this sense, we regard both Handel’s physical intuition and the subsequent criticisms of Refs. [12,13,14] as indispensable contributions: together they mapped out the logical constraints that a correct theory must satisfy. The relation to Handel’s theory and to its criticisms is discussed in detail in Appendix F. A complementary, phenomenological route to the same 1 / f current noise—through the amplitude modulation (beat) of electron wave packets dressed by soft photons—was developed in [15,16]; the present paper provides the field-theoretic foundation for that picture.
A recent no-go objection, and the scope it clarifies. Fukuyama [17] emphasizes two important facts: the infrared sector of ordinary QED scattering is coherently dressed and infrared-safe, and the free, source-free Maxwell field in four-dimensional de Sitter space has no secular infrared growth. We adopt the first statement as the QED benchmark rather than as an opposing description. FK dressed states give the same infrared-safe scattering physics as the Bloch–Nordsieck inclusive construction [18,19]. Our question is the real-time continuation: a current undergoing repeated scattering does not carry one immutable soft profile but a history α [ J ( t ) ] . The CTP influence functional describes this history, and its symmetric part supplies the stochastic spread of the evolving dressing. Thus, the itinerant vacuum is the dynamical extension of the standard coherent dressing, not an alternative S-matrix prescription. The second statement remains our “dry-noise” benchmark: without interaction-generated retarded dynamics, a formal stochastic representation does not by itself constitute classical Langevin evolution. Section 5 summarizes this relation.
Relation to prior work, and the origin of the present approach. The two ideas that distinguish this paper have a long history, beginning in the author’s own early work. A Langevin equation for the mean field was derived directly from the CTP effective action in 1986 [20], and the random force was shown to be, in general, non-Gaussian and colored. That derivation treated the mean field—the vacuum expectation value of the order parameter—directly, without a system–environment split. It was applied to the origin of density fluctuations in de Sitter space in [21] and to inflation as a quantum phase transition in [22]. The separation of an infrared-divergent imaginary kernel as the specific origin of the stochastic force was then made explicit in the inflationary context in [23] and developed into the dynamical phase-transition picture of [24]. The present paper unifies and extends this line of work.
It is useful to contrast this with the influential and largely parallel program of Calzetta, Hu, and collaborators [25,26,27]. That program obtains a Langevin equation by coarse graining: one first chooses a system–environment split (or a long-wavelength/short-wavelength division), traces out the environment, and arrives at an influence functional whose imaginary part supplies the noise. Coarse graining is the engine of the construction. Both routes yield a fluctuation–dissipation structure and are physically valuable. The distinctive feature of the present approach—implicit already in [20] and made sharp for infrared divergences here—is that the stochastic source arises from separating an infrared-divergent imaginary kernel out of the deterministic effective action, an operation that is mandatory rather than optional precisely because the kernel diverges—in the sense made precise in Section 2: not because a divergent force appears, but because the IR-enhanced variance invalidates the mean-field-plus-perturbation description. Constructing the reduced effective action already entails identifying the collective variable and integrating out unresolved degrees of freedom; the subsequent Hubbard–Stratonovich separation requires no additional Wilsonian coarse-graining step. Coarse graining determines which variable to follow; it does not create the stochastic source. This distinction is what allows the same mechanism to operate in QED, gravity, de Sitter space, and phase transitions on an equal footing, and it is what connects to the analysis of infrared quantum information in [28,29,30,31].
The new contributions beyond [23,24] are fourfold. The first is a systematic CTP derivation of the 1 / f noise spectrum for conserved electric currents: the gauge-invariant current-current influence functional for a dilute assembly of charged carriers generates a current noise S J ( ω ) = A J / | ω | whose amplitude A J ∝ n e 2 C IR / γ imp 2 , with C IR being the per-carrier soft-photon force coefficient, is fixed by the carrier density n, charge e, velocity β , and impurity scattering rate γ imp , up to the material matching factors specified in the soft kernel. This revisits Handel’s infrared QED proposal (discussed above), now as the proper CTP identity, and resolves the orthogonality objections of Refs. [12,13,14]: soft-sector orthogonality is reduced-state decoherence, and the missing rigorous bridge is the chain (1). The second is a parallel treatment of soft gravitons, connecting the soft-graviton IRD to metric noise and gravitational memory through the Einstein–Langevin equation. The third is a CTP treatment of non-equilibrium phase transitions, in which the λ ϕ 4 interaction generates multiplicative noise (a field-dependent noise kernel generated at loop level; Section 9), producing non-Gaussian tails in the curvature perturbation that are a potential source of primordial black hole overproduction [32,33,34]. The fourth is a unified presentation of all four arenas within a single CTP framework, clarifying the role of Ward identities, diffeomorphism invariance, and the distinction between dry and non-dry noise.
The itinerant vacuum: a unifying perspective. The four arenas treated in this paper share a deeper structural feature that motivates the title. We first fix the terminology. Throughout this paper “vacuum” does not mean that the Poincaré-invariant ground state of the full theory changes; in QED that state remains simple. It means the infrared-dressed sector—the state of the soft modes—that accompanies a given value of the collective variable (a current, a long-wavelength field, or an asymptotic charge). The family of such sectors is what the collective variable labels, and what changes along a history is the dressing, not the underlying ground state. In each case, the Langevin equation derived from the CTP effective action describes the time evolution of a field expectation value—a collective classical variable that parameterizes a continuous family of inequivalent quantum vacua. As the stochastic noise drives the order parameter through its configuration space, the system traverses a succession of distinct quantum vacua:
  • QED: the dressed coherent states | γ [ p ] ⟩ soft of Kulish–Faddeev [6] and Kibble [5] constitute a family of vacua labeled by the charged-particle momentum p. Each scattering event shifts the vacuum from | γ [ p ] ⟩ to | γ [ p ′ ] ⟩ . These are orthogonal: soft ⟨ γ [ p ′ ] | γ [ p ] ⟩ → 0 in the infrared limit. (The precise dress code is fixed by the conservation law of the asymptotic symmetry, and the original Kulish–Faddeev dressing requires refinement for a fully infrared-finite S-matrix [35,36,37]; this refinement sharpens, rather than alters, the vacuum-family picture used here.)
  • De Sitter inflation: the inflaton coherent vacuum | vac [ ϕ 0 ] ⟩ shifts with the field expectation value ϕ 0 ( t ) . The stochastic kicks of the noise ξ ( t ) randomly displace the vacuum along the potential, giving rise to the scale-invariant spectrum and, through the interaction-generated multiplicative noise, to rare large displacements that can form primordial black holes.
  • Phase transitions: the order-parameter manifold { | θ ⟩ } is the vacuum family. The Langevin equation is a stochastic geodesic on this manifold. Domain walls are boundaries between regions that have wandered to different vacua.
  • Gravity: the BMS vacuum [38] is not unique; an infinite-dimensional family of vacua related by BMS supertranslations is degenerate. A burst of gravitational radiation is a transition between BMS vacua—this is precisely the statement of Strominger and Zhiboedov [39] that gravitational memory is the transition between BMS vacua and that the Fourier transform of the memory formula reproduces Weinberg’s soft theorem. Gravitational memory is the classical remnant of this vacuum shift, encoded in the ω → 0 limit of the retarded kernel.
In all four cases, the infrared divergence is the field-theoretic cost of vacuum itinerancy. The orthogonality of soft states—the central objection to Handel’s theory—is not a pathology of the formalism but the precise statement that different classical histories correspond to orthogonal quantum vacua. The retarded kernel encodes the memory of past vacuum transitions. The noise kernel encodes the quantum uncertainty of which vacuum will be occupied next. The Hubbard–Stratonovich identity converts this quantum uncertainty into a classical stochastic force, without any coarse graining beyond the reduction to the collective-variable effective action and without a separate decoherence postulate. Removing the IRD from the deterministic action would erase the physics of vacuum itinerancy entirely; the correct response is to relocate it into the noise.
Convergence with current infrared physics. The viewpoint of this paper, while not the textbook treatment of infrared divergences, is far from isolated. It converges with several independent lines of recent development. First, the modern infrared triangle of Strominger and collaborators [38,39,40] establishes the equivalence of asymptotic symmetries, soft theorems, and memory effects, and identifies gravitational (and electromagnetic) memory as a transition between degenerate vacua—just the itinerant-vacuum picture, reached there through the S-matrix and celestial-holography route rather than the real-time CTP route used here. Second, the demonstration by Danielson, Satishchandran, and Wald [41] that horizons decohere quantum superpositions through the emission of soft photons and gravitons that imprint a which-path memory shows that soft radiation carries genuine, physical decoherence—the same mechanism that, in our language, supplies the noise kernel. Third, the program of infrared-finite scattering theory [42] and of infrared quantum information [28,29,30,31] treats the soft sector as carrying physical information rather than as a mere technical nuisance. In the same spirit, Hirai and Sugishita [35,36,37] have shown that the dress code required for an infrared-safe S-matrix is fixed by the conservation law of the asymptotic symmetry—which is nothing but the memory effect—so that the correct dressing is itself a statement about transitions among the degenerate vacua of the asymptotic symmetry. These developments approach the infrared sector from the perspective of real-time, non-perturbative, and memory-carrying dynamics, and are complementary to the inclusive S-matrix perspective in which the divergence simply cancels. The present paper belongs to this real-time tradition and adds to it a direct stochastic representation and a concrete condensed-matter prediction.
Structure of the paper. Section 2 presents the CTP effective action and the algebraic stochastic separation. Section 3 specializes to 1 / f noise and clarifies time-scale conventions. Section 4 treats soft QED in detail. Section 5 relates the construction to the standard Faddeev–Kulish dressing, shows that the Bloch–Nordsieck/KLN cancellation is built into the CTP influence functional, and delimits the scope of the stochastic interpretation. Section 6 treats soft gravitons, metric noise, and gravitational memory. Section 7 treats non-equilibrium phase transitions. Section 8 summarizes the de Sitter scalar case. Section 9 discusses higher loops and non-Gaussian noise. Section 10 collects observable consequences and falsifiability tests. Section 11 compares the construction with coarse-graining and decoherence approaches. Appendix A records the Hubbard–Stratonovich identity. Appendix B collects formulae for the 1 / f kernel. Appendix C derives the Wilson-line form of the QED influence functional. Appendix D gives the parallel soft-graviton derivation and relates the temporal 1 / f and spatial 1 / k 3 spectra. Appendix E gives the scaling estimate for the soft-photon noise coefficient. Appendix F discusses the relation to Handel’s theory. Appendix G records the de Sitter noise kernel. Appendix H briefly compares itinerant vacua with coherent resonating states.

2. Closed-Time-Path Effective Action and Exact Stochastic Separation

We use the real-time closed-time-path and influence-functional formalism [26,43,44,45,46,47]. Let q denote the collective variable whose infrared sector is to be followed. It may be the trajectory of a charged particle, an electric current, a long-wavelength scalar field, a metric perturbation, or an order parameter. In all cases q parameterizes a family of quantum vacua: each value q 0 carries a coherent vacuum | vac [ q 0 ] ⟩  (an infrared-dressed sector in the sense fixed in Section 1), and the Langevin equation to be derived below is the stochastic equation of motion for this itinerant vacuum. For a fundamental field q, the following path integral and the bare action S [ q ] are literal. For a composite observable such as the QED current j μ = e ψ ¯ γ μ ψ , they are schematic: one must retain the underlying matter fields in the blocked action, or construct a separate composite-operator effective action [48,49]. The CTP generating functional has two histories, q + and q − ,
Z [ J + , J − ] = ∫ D q + D q − exp i S [ q + ] − i S [ q − ] + i ∫ ( J + q + − J − q − ) .
In Keldysh variables,
q r = q + + q − 2 , q a = q + − q − ,
after integrating out the microscopic or infrared fields, the CTP effective action takes the form
Γ [ q r , q a ] = Γ R [ q r , q a ] + i 2 ∫ d x d x ′ q a ( x ) N ( x , x ′ ) q a ( x ′ ) + Γ ng [ q r , q a ] .
Γ R is real and causal. N is the symmetric positive Hadamard (noise) kernel. Γ ng contains higher powers of q a corresponding to non-Gaussian noise. To linear order in q a —the order that determines the equation of motion—the causal part has the generic structure
Γ R [ q r , q a ] = ∫ d x q a ( x ) E [ q r ] ( x ) + ∫ d x d x ′ q a ( x ) Σ R ( x , x ′ ) q r ( x ′ ) + O ( q a 2 , q a 3 ) ,
where E [ q r ] ( x ) ≡ δ S [ q ] / δ q ( x ) | q = q r is the local (tree-level) classical equation-of-motion operator obtained from the bare action S—for a scalar field, E [ q ] = ( □ + m 2 ) q + V ′ ( q ) —and Σ R is the retarded self-energy kernel generated by integrating out the microscopic modes. The split between E and Σ R is understood after renormalization: the radiative corrections contained in the O ( q a ) term have a local part, which is absorbed into the renormalized parameters (mass, charge, and couplings) of E, and a nonlocal remainder, which is what we denote by Σ R . Thus, E is the classical equation of motion with physical parameters, while all memory and dissipation reside in Σ R ; in soft QED, the latter contains the soft virtual-photon contribution (Section 5). Thus, E [ j ] = δ S [ j ] / δ j is not assumed for QED: the microscopic equations follow by varying ψ and ψ ¯ , including the chain-rule variation in the current-dependent influence functional. The path weight contains
exp i Γ R [ q r , q a ] − 1 2 ∫ q a N q a + i Γ ng .
The ordinary Gaussian Hubbard–Stratonovich identity applies to the quadratic q a N q a term. Higher pure- q a vertices are treated in Section 9 through the generalized characteristic functional and non-Gaussian cumulants; a genuine classical probability functional requires the positivity of the full characteristic functional.
Status of the O ( q a 3 ) terms. Terms of third and higher order in q a do not admit a Gaussian Hubbard–Stratonovich representation, so their relevance must be assessed separately. Two facts control them. (i) Soft QED. The photon field is free and couples linearly to the conserved current. Integrating it out, therefore, yields an influence action that is exactly quadratic in j a [Equation (20) below; Feynman–Vernon [45]]. The soft photon sector generates no O ( j a 3 ) term at all, and the Gaussian identity (7) is exact there. Terms cubic in q a arise only from the nonlinearity of the charged matter itself (recoil). These correspond to subleading soft factors; they are infrared finite and do not carry the ∫ d ω / ω enhancement. (ii) Generic interacting theories ( λ ϕ 4 , gravity). Restoring ℏ, the CTP weight is exp { ( i / ℏ ) Γ } , and the relevant width of q a is of order ℏ. Writing q a = ℏ q ¯ a , the term linear in q a is ℏ-independent (classical drift), the quadratic term is ℏ q ¯ a N q ¯ a , which grows with occupation through (11), and a term of order q a 3 carries an extra ℏ 2 with no occupation enhancement. In the infrared-enhanced regime the cubic vertex is, therefore, suppressed relative to the noise term by ( ρ / F ) 2 ∼ 1 / n 2 , where F and ρ are the statistical and spectral functions. This is the standard criterion of the classical-statistical approximation [50]. In generic theories the classical stochastic reading is, therefore, a controlled approximation, not an identity. It holds in the infrared-enhanced regime, and its corrections can be computed perturbatively (Section 9).
If N has an infrared divergence, leaving it inside the path integral as a divergent quadratic coefficient obscures the physics. The Hubbard–Stratonovich identity [51,52]
exp − 1 2 ∫ q a N q a = ∫ D ξ P [ ξ ] exp i ∫ d x q a ( x ) ξ ( x )
with
P [ ξ ] = N exp − 1 2 ∫ d x d x ′ ξ ( x ) N − 1 ( x , x ′ ) ξ ( x ′ ) , ⟨ ξ ( x ) ξ ( x ′ ) ⟩ ξ = N ( x , x ′ )
is an identity of the generating functional. The Langevin equation follows by varying Γ R in (5) with respect to q a and setting q a = 0 :
E [ q ] ( x ) + ∫ d x ′ Σ R ( x , x ′ ) q ( x ′ ) = − ξ ( x ) ,
where E [ q ] is the local equation-of-motion operator and Σ R the retarded (causal) self-energy kernel of (5). Equation (9) is a closed equation only when an appropriate effective action and dynamical closure for q have been established; in particular, variation in the QED influence functional with respect to a composite current alone does not supply its matter dynamics. In this reduced real-time representation, with the usual infrared-safe asymptotic dressing understood, the infrared-enhanced symmetric contribution relevant to stochastic evolution resides in the covariance N. This statement does not remove the coherent soft dressing or causal memory contained in the retarded sector; rather, it identifies the part of the same infrared physics that becomes a stochastic spread of real-time histories.
The stochastic source ξ is classical in the sense that the CTP functional is represented as an average over c-number histories. This is weaker and more fundamental than wave-function collapse: it is a representation of the in–in generating functional. It is worth emphasizing why this classical representation is natural rather than forced. Unlike an ultraviolet divergence, which is intrinsically quantum (it requires loops, i.e., powers of ℏ), an infrared divergence does not depend on ℏ and survives the classical limit: classical bremsstrahlung already carries the same ∫ d ω / ω soft divergence, with the same velocity factor, as the quantum soft-photon emission (Low’s theorem [53] guarantees the leading soft radiation is classical). Representing an IRD as a classical stochastic field is, therefore, not an operation that forces a quantum object into a classical mould; it extracts the classical side of a divergence that straddles the quantum and classical descriptions from the outset. The imaginary part of the influence action also suppresses histories with large q a , providing reduced-state decoherence [26,47]:
ρ red [ q + , q − ] ∝ exp − 1 2 ∫ q a N q a .
Decoherence and stochasticity arise from the same imaginary influence action; neither requires a separate postulate.
When is the classical reading faithful? The occupation criterion. A caveat must be stated with equal care, because the Hadamard kernel is not special to infrared-divergent systems: N = ⟨ { q , q } ⟩ is finite and nonzero in any quantum system whatsoever, down to the vacuum zero-point fluctuations of a harmonic oscillator. The identity (7) can be applied there too, and it would be a misreading to conclude that every quantum fluctuation is, thereby, a classical statistical one. The sharp criterion is the ratio of the symmetric to the antisymmetric part of the two-point function,
⟨ { q , q } ⟩ ω ⟨ [ q , q ] ⟩ ω = 2 n ( ω ) + 1 → in case of thermal equilibrium coth β ω 2 ,
where n ( ω ) is the occupation number of the mode. The commutator (spectral function) is fixed by the canonical algebra and does not grow; the anticommutator grows with occupation. The classical stochastic representation reproduces the symmetric correlators exactly, at any n, but observables that depend on non-commutativity beyond linear response—genuine interference, Leggett–Garg-type tests—are missed, with a relative error of order 1 / n . For an ordinary vacuum mode, n ≃ 0 and the ratio (11) is of order one: the Hubbard–Stratonovich field is then a bookkeeping device—the “quantum noise” of zero-temperature quantum Brownian motion [46]—and must not be interpreted as classical statistics. The soft limit requires one further distinction. In an FK cloud, the infrared logarithm ⟨ N γ ⟩ ∼ ∫ d ω / ω is a cumulative sum over infinitely many soft modes; it does not imply that every individual mode has a macroscopic occupation. The more robust classicality/decoherence criterion is, therefore, the cumulative suppression of overlaps between differently dressed histories, | ⟨ α 2 | α 1 ⟩ | = e − D 12 , together with anticommutator dominance where a mode description is meaningful. Squeezed super-horizon modes and thermal modes can, additionally, have n k ≫ 1 . Thus, the Hubbard–Stratonovich identity does not by itself classicalize quantum mechanics; the classical statistical reading becomes accurate where the infrared accumulation makes alternative histories effectively distinguishable and the symmetric fluctuations dominate.
The criterion has an equivalent algebraic and operational statement. Normal ordering removes precisely the vacuum unit from the ratio (11): mode by mode, the Hadamard kernel splits as ( 2 n + 1 ) ρ = ρ + 2 n ρ , and since the zero-point part ρ and the normal-ordered part 2 n ρ are separately positive and symmetric, the identity (7) may be applied to either term. The normal-ordered part is what an absorptive photodetector measures [54]; equivalently, it is the component of the quantum noise spectrum symmetric under ω → − ω and, hence, shareable with a classical detector, the excess ρ being the emission–absorption asymmetry ( n + 1 versus n) that a ground-state detector cannot supply [55]. A hybrid scheme that converts only 2 n ρ into a classical source while retaining ρ as a genuine quantum influence functional [46] differs from the full transformation by the same relative order 1 / n , so in the infrared-divergent sector the scheme dependence vanishes and the classical sector is delimited without ambiguity. In an interacting theory or a curved spacetime, where no global mode decomposition is available, the local covariant version of this split is Hadamard point-splitting subtraction [56,57]: the subtracted Hadamard parametrix is universal and state-independent and carries the ultraviolet vacuum fluctuations, whereas the infrared divergence resides entirely in the smooth, state-dependent remainder—a renormalization-theoretic restatement of the claim that an infrared divergence is statistical rather than pathological.
The classical–quantum boundary is frequency dependent. By (10), two histories that differ by q a are distinguished with weight e − D [ q a ] , where
D [ q a ] = 1 2 ∫ d ω 2 π | q a ( ω ) | 2 N ( ω ) .
For an infrared-enhanced N, D accumulates from the low-frequency end. The scale ω c at which the accumulated D for the resolved differences reaches order unity separates two regimes. Below ω c the soft sector makes alternative histories effectively distinguishable, and it is represented by the stochastic source of the pointer variable q. Above ω c it does not decohere them, and its effect is absorbed into the renormalized parameters of E and the dispersive part of Σ R . This is the real-time counterpart of the renormalization-group picture, in which infrared contributions pile up as the scale is lowered and the boundary is the scale below which they can no longer be hidden in the macroscopic parameters of the pointer variable. The boundary ω c depends on the source strength (through q a ) and on the observation time (through the lower end of the frequency integral).
Diagrammatic meaning of N: a real-time cutting rule. The pair ( Σ R , N ) in (4) and (5) is not an ad hoc split but the real-time (closed-time-path) generalization of the ordinary Cutkosky unitarity cut [58]. In the in–out formalism, the unitarity of the S-matrix relates the imaginary part of a forward amplitude to a phase-space integral over on-shell intermediate states (the optical theorem); Cutkosky’s rule computes this diagrammatically by replacing internal propagators with their on-shell delta functions. The closed-time-path (or finite-temperature real-time) formalism extends this cutting prescription to off-forward and non-equilibrium correlators: every internal line of a diagram can be cut in all possible ways, with the cut propagator replaced by its Wightman (positive/negative-frequency) function, and the sum over such cuts reconstructs the imaginary, symmetric part of any CTP vertex or self-energy [59,60,61,62]. In this language, Σ R is the uncut (dispersive) diagram and N is the same diagram with its internal line(s) cut on shell; in thermal equilibrium this reduces to the familiar fluctuation–dissipation relation N ( ω ) = coth ( β ω / 2 ) Im Σ R ( ω ) . Section 4 and Section 6 exhibit this explicitly: N QED μ ν and N μ ν α β are the on-shell cuts, 2 π δ ( k 2 ) , of the single-photon and single-graviton exchange diagrams between two current or stress-tensor insertions, while the sunset noise kernel of Section 7 is the simultaneous triple cut of the two-loop self-energy diagram generated by the quartic self-coupling. The cutting-rule reading does not add any new assumption; it is simply the statement, made diagrammatic, that a positive symmetric kernel obtained from unitarity is directly what the Hubbard–Stratonovich identity converts into a classical noise covariance.
Why the separation is mandatory, not optional. In the standard route of Refs. [25,26,27], a system–environment split is chosen first; tracing over the environment then produces the influence functional. Coarse graining precedes the stochastic representation. The present approach inverts this logic. Given any CTP effective action whose imaginary part is a positive symmetric kernel, the Hubbard–Stratonovich identity (7) applies as an algebraic identity—regardless of whether a natural “environment” has been identified.
It is worth being precise about where an infrared-divergent N does its damage because the answer is not the naive one. Since the N term in (4) is quadratic in q a , it drops out of the mean-field equation of motion obtained by varying with respect to q a and setting q a = 0 : the equation for ⟨ q ⟩ never contains N directly, divergent or not. (In the in–out formalism the situation is worse: there, the equation of motion involves the time-ordered self-energy Σ F = Σ R + ( correlation part ) , and the divergent Hadamard part does contaminate the equation as a complex, acausal term—one classic failure mode.) In the in–in formalism the divergence instead strikes through two indirect channels. First, in any interacting theory the Hadamard function feeds back into the deterministic kernels through loops: the tadpole δ m 2 ∼ λ ∫ G IR H ∼ λ ∫ d ω / ω and its higher-loop relatives are infrared divergent, so the perturbative expansion of Σ R and U eff around the mean field is destroyed by secular infrared logarithms—the familiar secular-growth problem of de Sitter and critical dynamics. Second, even where the mean field itself remains finite and regular, the divergence of N means that the variance ⟨ q 2 ⟩ − ⟨ q ⟩ 2 diverges: the mean ceases to represent any individual realization. In a symmetry-breaking transition ⟨ Φ ⟩ stays at the symmetric point forever—a correct ensemble average that describes no single history, while every actual sample or universe falls into one broken vacuum.
The Hubbard–Stratonovich separation addresses both failures at once, and this is the precise sense in which it is mandatory rather than optional. It is not a device for removing a divergent force (no such force exists in the in–in equation); it is a reorganization of the theory—an exact resummation. The description shifts from “mean field plus perturbative quantum fluctuations around it,” which the infrared divergence renders useless, to “an ensemble of classical stochastic histories,” each of which is finite and physical, with the divergence relocated into the spread of the ensemble. Correlators computed from the Langevin dynamics are finite and self-consistently regulated by the dissipation and the nonlinearity—the Starobinsky–Yokoyama equilibrium distribution [8] being the canonical example, known to reproduce the resummation of the leading infrared logarithms to all orders. The identity (7) guarantees consistency (no double counting); the physics it unlocks is non-perturbative resummation and, equally important, access to single-history questions: a physical measurement is one realization, not an ensemble average, and only the stochastic representation describes what one realization does.
Why no additional coarse graining is needed: self-selection of the order parameter. The reason the stochastic representation requires no additional coarse graining is worth stating once as a principle, since it recurs in every arena below. In a system with an instability or an infrared divergence, a macroscopic order parameter q emerges dynamically of its own accord: the unstable or infrared-enhanced long-wavelength mode grows, becomes classical, and comes to dominate and characterize the whole system. (In soft QED, no individual soft mode acquires a macroscopic occupation. What becomes macroscopic is the collective current, together with the cumulative dressing it drags; see the overlap criterion above.) The remaining microscopic degrees of freedom do not organize into structure; they act as an effective environment for q whether or not one chooses to call them so. The decomposition “system = order parameter + effective environment” is, thus, not an external modeling choice imposed by the physicist, but a structure the dynamics selects by itself—one is compelled to follow q because it is what becomes macroscopic and visible. Coarse graining, in the usual sense, only re-describes after the fact a split that the instability has already made. This is why the same Hubbard–Stratonovich identity operates identically for soft QED, soft gravitons, de Sitter fields, and phase transitions: in each, an infrared or unstable sector self-selects the order parameter, and the rest supplies the noise. Throughout this paper, when we say that the construction needs no additional coarse graining, this self-selection (on top of the reduction to Γ [ q ] ) is what is meant.
The original infrared divergence is not erased by the separation; it is relocated into the stochastic covariance. For a 1 / f kernel, this relocation is explicit: the equal-time variance
⟨ ξ 2 ⟩ = A π ln ω max ω IR → ω IR ∼ T obs − 1 A π ln ( ω max T obs )
grows logarithmically with observation time T obs . The − 1 power law persists toward lower frequency as T obs increases and diverges only in the ideal limit T obs → ∞ . This is the precise “price” of the original IRD: not an infinite deterministic force, but an indefinitely accumulating stochastic variance that is finite and well-defined at every finite observation time.

3. The 1 / f Infrared Sector

A universal long-memory kernel is obtained when the noise spectrum has the form
N 1 / f ( ω ) = A | ω | Θ ( | ω | − ω IR ) Θ ( ω max − | ω | ) .
The integrated variance is
⟨ ξ 2 ⟩ = ∫ d ω 2 π N 1 / f ( ω ) = A π ln ω max ω IR ,
which is the original IRD now residing in the stochastic covariance rather than in a deterministic coefficient.
For an observed variable q with linear response function χ ( ω ) (e.g., a Drude conductivity), the observable power spectrum is
S q ( ω ) = | χ ( ω ) | 2 N 1 / f ( ω ) .
If | χ ( ω ) | 2 is approximately constant at low frequency, S q ( ω ) ∝ 1 / | ω | , giving 1 / f noise directly.
Two time scales must be distinguished. The upper frequency ω max fixes the shortest time in the infrared scaling regime,
t UV = ω max − 1 .
The total observation time fixes the lower cutoff,
t IR = ω IR − 1 ∼ T obs .
Increasing T obs lowers ω IR and enlarges the logarithm (15) while the − 1 power law itself remains. This is the precise sense in which the original IRD survives: not as an infinite deterministic force, but as the indefinitely extensible long-memory part of the stochastic source.

4. Soft QED: 1 / f Current Noise and Source-Parameter Dependence

4.1. Gauge-Invariant Influence Functional

The QED soft sector is the cleanest arena because the photon field is Gaussian before coupling to charged matter, while the leading soft corrections are constrained by Ward identities and exponentiate.
Consider a macroscopic conserved current built from N charged carriers, scattered, for example, by impurities, with charge e, carrier density n, and characteristic velocity β = v / c :
j μ ( x ) = e ∑ i = 1 N ∫ d τ i u i μ ( τ i ) δ ( 4 ) ( x − z i ( τ i ) ) , ∂ μ j μ = 0 .
Integrating out the photon field on the CTP contour gives the influence action
Γ γ [ j r , j a ] = − ∫ d 4 x d 4 x ′ j a μ ( x ) D μ ν R ( x − x ′ ) j r ν ( x ′ ) + i 2 ∫ d 4 x d 4 x ′ j a μ ( x ) D μ ν H ( x − x ′ ) j a ν ( x ′ ) .
More precisely, Equation (20) is part of the blocked CTP action S block [ ψ ¯ ± , ψ ± ] = S m [ ψ ¯ + , ψ + ] − S m [ ψ ¯ − , ψ − ] + Γ γ [ j + , j − ] , where j ± μ = e ψ ¯ ± γ μ ψ ± are constrained composite functionals, not independent elementary fields. This blocked action is distinct from the Legendre-transformed 1PI effective action. Variation in the fermions (or their CTP Green functions) retains both their matter dynamics and the chain-rule contributions from Γ γ . A direct variation with respect to j a identifies the photon-induced retarded and stochastic terms, but is not by itself a current equation of motion. Because the free photon field is Gaussian and couples linearly to j μ , (20) is exact for a given current history. Integrating out the photons generates no term of higher order in j a (Section 2). The Hadamard function
D μ ν H ( x − x ′ ) = ⟨ { A μ ( x ) , A ν ( x ′ ) } ⟩ = ∫ d 4 k ( 2 π ) 4 2 π δ ( k 2 ) P μ ν T ( k ) e − i k ( x − x ′ )
is positive on physical transverse modes or, equivalently, on conserved currents. Here, P μ ν T ( k ) is the transverse polarization sum: in a physical (radiation) gauge with a timelike reference vector n μ it reads
P μ ν T ( k ) = ∑ λ = 1 , 2 ϵ μ ( λ ) ( k ) ϵ ν ( λ ) * ( k ) = − η μ ν + k μ n ν + k ν n μ k · n − k μ k ν ( k · n ) 2 ,
which satisfies k μ P μ ν T = 0 on shell; when contracted with a conserved current ( k μ j μ = 0 ), only the spatial transverse part δ i j − k ^ i k ^ j survives, so the gauge-dependent terms drop out. The gauge-invariant form of (20) is a CTP Wilson-line expectation value (see Appendix C), which automatically combines self-energy and vertex infrared contributions. By the Ward–Takahashi identity [63,64]
k μ Γ μ ( p + k , p ) = S − 1 ( p + k ) − S − 1 ( p ) ,
where Γ μ ( p + k , p ) is the full electron–photon vertex function and S ( p ) is the full electron propagator (so that S − 1 ( p ) is the inverse propagator), the vertex and self-energy IRDs are not independent noise channels; they constitute a single gauge-invariant object.
Diagrammatic origin: a cut single-photon exchange. Diagrammatically, (20) is the diagram in which one photon line connects two insertions of the current j μ ; D μ ν R is this diagram evaluated in the ordinary (uncut) sense, and D μ ν H in (21) is the same diagram with the internal photon line cut on shell—the real-time cutting rule of the general discussion in Section 2, here reducing to the familiar Bloch–Nordsieck/Yennie–Frautschi–Suura statement that the imaginary part of the virtual-photon exchange equals the phase-space integral over real soft-photon emission. Cutting a bare one-loop electron self-energy diagram alone would not be gauge invariant; consistency requires including the cut vertex diagrams as well, which is the content of the Ward–Takahashi identity (23). The Wilson-line representation of Appendix C sidesteps this bookkeeping: since it contains only insertions along a single worldline, with no distinction between “self-energy” and “vertex” attachment points, its cut is automatically the gauge-invariant sum of all such contributions to every order, and reduces to (21) directly (see Appendix C for the explicit consistency check between the resummed exponent and (23)).

4.2. Noise Kernel and 1 / f Spectrum

The Hubbard–Stratonovich identity applied to (20) produces a c-number stochastic gauge potential a μ with ⟨ a μ ( x ) a ν ( x ′ ) ⟩ = D μ ν H ( x − x ′ ) . For the current (19), the current-projected noise kernel in the soft limit ( | k | → 0 ) follows from the on-shell photon phase space,
∫ d 3 k ( 2 π ) 3 2 | k | ( p · ϵ ) 2 ( p · k ) 2 ∝ ∫ ω IR ω max d ω ω f ( β ) ,
with a velocity form factor such as
f ( β ) = 1 β ln 1 + β 1 − β − 2 → β → 0 2 3 β 2 , f ( β ) → β → 1 ∞ .
For N = n V identical sources with isotropic velocity distribution, the angular integral and sum over sources give the noise kernel
N QED μ ν ( ω ) = n e 2 f ( β ) | ω | P μ ν T Θ ( | ω | − ω IR ) Θ ( ω max − | ω | )
(see Appendix C for the detailed derivation).
This is a central result. The noise spectrum is N ( ω ) ∝ 1 / | ω | , i.e., 1 / f , arising purely from the soft-photon phase space. The amplitude is determined entirely by Ward-identity-constrained source parameters: n, e, β . In the non-relativistic limit β ≪ 1 , the amplitude scales as n e 2 β 2 .
It is worth being explicit about what N QED μ ν represents physically, since three related but distinct two-point objects appear in this section and the next. The Hubbard–Stratonovich field a μ has the correlator ⟨ a μ a ν ⟩ = D μ ν H (21): this is the statistical correlation of the random gauge field at the level of the photon field itself, computed here in a vacuum, without any scattering medium. Eikonal projection onto prescribed charged trajectories yields the soft-dressing weight N QED μ ν in Equation (26). This weight is not automatically the power spectrum of the Lorentz force e F μ ν [ a ] u ν : the field strength involves derivatives of the HS gauge potential, and a physical current spectrum also requires the matter response. Identifying the transport force covariance with a 1 / | ω | kernel is, therefore, a matching assumption used in Section 4.3, to be checked microscopically. Schematically,
D μ ν H → eikonal projection N QED μ ν → transport matching and response S J ( ω ) .

4.3. Stochastic Equation of Motion and Impurity-Density Dependence

We now adopt a phenomenological transport closure to connect the soft-photon influence functional to current fluctuations. This is an effective approximation, not a derivation of an autonomous current equation directly from QED. For an impurity-dominated conductor, the drift velocity of a charge carrier obeys the Drude–Langevin equation
m v ˙ ( t ) + m γ imp v ( t ) = e E + ξ IR ( t ) + ξ th ( t ) ,
where ξ th is thermal (Johnson–Nyquist) noise and
γ imp = τ imp − 1 ≃ n imp v F σ tr
is the impurity scattering rate. The infrared force ξ IR is the soft-photon stochastic source from Section 4.2 acting on a single carrier, i.e., the per-carrier ( N QED μ ν / n ) scalar projection of N QED μ ν onto the current direction; its stationary correlation is
⟨ ξ IR ( t ) ξ IR ( t ′ ) ⟩ = ∫ d ω 2 π e − i ω ( t − t ′ ) N IR ( ω ) , N IR ( ω ) = C IR ( n imp ) | ω | ,
Here, the 1 / | ω | form of the force kernel is a model-dependent matching ansatz motivated by the eikonal soft factor in (26) rather than a consequence of the HS identity alone; Appendix E estimates C IR under this assumption.
We want the power spectral density of the electric current. For N c = n V carriers in a volume V, the current density is J = − ( e / V ) ∑ i v i . The soft-photon forces on different carriers are uncorrelated since each carrier’s dressing is updated by its own scattering events. We define the spectral density per unit volume, S J ( ω ) , in the stationary state by the Wiener–Khinchin relation
⟨ δ J ( t ) δ J ( t ′ ) ⟩ = 1 V ∫ d ω 2 π e − i ω ( t − t ′ ) S J ( ω ) , S J ( ω ) = n e 2 S v ( ω ) ,
where S v ( ω ) is the single-carrier velocity spectral density and δ J = − ( e / V ) ∑ i δ v i is the fluctuating part of the current. Solving (28) in frequency space for the fluctuating velocity (dropping the deterministic drive E, which sets the mean current rather than its fluctuations),
δ v ( ω ) = ξ IR ( ω ) + ξ th ( ω ) m ( γ imp − i ω ) ,
the response is the Drude (Lorentzian) susceptibility χ ( ω ) = [ m ( γ imp − i ω ) ] − 1 , with | χ ( ω ) | 2 = [ m 2 ( γ imp 2 + ω 2 ) ] − 1 . Taking the modulus squared of (32) and using the stationarity and independence of the two noises ( ξ IR originates in the soft photon field and ξ th in the electron–phonon/impurity bath; their kernels are positive and additive, so exp [ − 1 2 q a ( N IR + N th ) q a ] factorizes and the Hubbard–Stratonovich identity (7) incorporates both at once as ξ IR + ξ th with ⟨ ξ IR ξ th ⟩ = 0 ),
S v ( ω ) = | χ ( ω ) | 2 N IR ( ω ) + N th ( ω ) = N IR ( ω ) + N th ( ω ) m 2 ( γ imp 2 + ω 2 ) .
Retaining the infrared part and evaluating it in the low-frequency band | ω | ≪ γ imp , where | χ ( ω ) | 2 ≃ ( m γ imp ) − 2 is flat, the current spectral density (31) becomes
S J IR ( ω ) = n e 2 | χ ( ω ) | 2 N IR ( ω ) ≃ n e 2 ( m γ imp ) 2 C IR ( n imp ) | ω | ≡ A J ( n imp ) | ω |
with the 1 / f noise coefficient
A J ( n imp ) = n e 2 ( m γ imp ) 2 C IR ( n imp ) .
Thus, the flat Drude response converts the 1 / | ω | force kernel directly into a 1 / f current spectrum; the Lorentzian rolloff at | ω | ≳ γ imp fixes the upper edge of the scaling band, ω max ∼ γ imp . If C IR ∝ n imp (one soft-dressing event per scattering) and γ imp ∝ n imp , then A J ∝ n imp − 1 . This is a falsifiable scaling prediction.
Comparing with the empirical Hooge formula [65],
S I ( f ) I 2 = α H N c f ,
the coefficient A J ∝ C IR / γ imp 2 of the absolute spectrum S J = A J / | ω |  can be estimated using the soft-sector input and the medium response; it is not yet a parameter-free first-principles determination of the phenomenological Hooge parameter. The independence of the carriers’ soft-photon forces is also what produces the 1 / N c factor in (36). For example, identifying the two spectra in the flat Drude band gives, up to the normalization by I 2 and the carrier number N c , α H ∼ C IR / γ imp p with an exponent p fixed by that normalization; the resulting dependence on impurity density and mobility is a falsifiable prediction, to be tested against samples with controlled n imp . We keep the absolute/relative distinction explicit because the power of γ imp in α H is sensitive to the I 2 and N c normalization, whereas the absolute-spectrum result A J ∝ C IR / γ imp 2 is not. The thermal contribution ξ th gives the standard Johnson–Nyquist white noise S J th ∝ T γ imp [66,67,68], which is distinct from and additive to the 1 / f term.
The memory interpretation is also direct: soft photons carry which-path information about charged-particle histories. The integrated 1 / f variance
⟨ δ J 2 ⟩ IR ≃ A J π ln ω max ω IR
is the electrical analogue of the inflationary variance ⟨ δ ϕ 2 ⟩ ≃ ( H / 2 π ) 2 ln ( k max / k min ) .
Phenomenological verification with electron wave packets. The abstract prediction of a 1 / f current noise from the soft-photon back reaction has been verified in a concrete, phenomenological wave-packet model in a companion study [16]. There, the electron is represented by a classical wave packet propagating in a (semi-)conductor and obeying a classical Langevin equation that includes the soft-photon back reaction; squaring the wave packet to form the electric current and demodulating the result yields a robust 1 / f spectrum. That treatment complements the field-theoretic derivation given here: the present paper fixes the noise amplitude and its parameter dependence from the CTP influence functional, while the wave-packet model exhibits the robustness and propagation of the resulting 1 / f signal and traces it through to thresholded, demodulated output. The two descriptions share the same physical origin—the soft-photon infrared sector acting back on the charged current—and are mutually consistent: the demodulation (beat) picture [15] and the influence-functional picture are complementary readings of the same infrared structure.

5. From Faddeev–Kulish Dressing to Itinerant Vacua

The standard QED infrared problem has two equivalent practical resolutions for physical scattering predictions. Bloch–Nordsieck/KLN sums over unresolved degenerate soft states, whereas the Faddeev–Kulish (FK) construction builds the soft cloud directly into the asymptotic charged states. In their common domain they give the same infrared-safe cross sections [5,6,18,19]. We, therefore, take FK dressing as the natural QED starting point.
For a fixed asymptotic scattering process, a charged current determines a multimode coherent profile, schematically
J ⟼ | α [ J ] ⟩ , α λ ( k ) ∝ e u · ϵ λ u · k ,
with the appropriate sum over incoming and outgoing charges. The dressed description absorbs the leading infrared structure into this coherent cloud. The additional step considered here is only dynamical: during repeated scattering the collective current evolves, so the dressing becomes a trajectory | α [ J ( t ) ] ⟩ . In this sense the “itinerant vacuum” is a continuously evolving family of infrared-dressed sectors; it is not a time-dependent bare Fock vacuum.
The CTP formulation keeps precisely the information discarded when one compresses the whole history into a single in–out event. Its symmetric kernel can be written by the Hubbard–Stratonovich identity as
N ⟶ ξ ( t ) ⟶ J ( t ) ⟶ α [ J ( t ) ] ,
whereas the retarded kernel determines how each current history is causally recorded in the field. Thus coherent dressing and stochastic itinerancy are not different infrared mechanisms: they are the asymptotic and real-time descriptions of the same current-driven soft sector. The unphysical infrared regulator is removed in the usual infrared-safe scattering construction; the finite-time scale ω IR ∼ T obs − 1 that appears below is, instead, a physical resolution scale of the real-time correlator.
The BN/KLN summation is built in, and the real-time result is infrared finite. This can be made explicit. The CTP influence functional is the trace over the photon field,
F [ j + , j − ] = Tr γ U j + ρ γ U j − † = e i Γ γ [ j r , j a ] ,
that is, a sum over all final soft-photon states. This trace is exactly the inclusive sum of Bloch–Nordsieck and Kinoshita–Lee–Nauenberg. In terms of the CTP components, the ( + + ) and ( − − ) propagators carry the virtual soft photons of each branch, and the ( + − ) Wightman function carries the real photons that cross the final-time cut. Unitarity gives F [ j , j ] = 1 , i.e.,
Γ γ [ j r , j a = 0 ] = 0
for every current history. On the diagonal, the virtual and real soft contributions cancel identically, and any photon-mass regulator drops out. This is the BN/KLN cancellation, now holding history by history. (In the language of almost-degenerate perturbation theory, the dressed basis | α [ J ] ⟩ is the appropriate zeroth-order basis, and | α [ J ( t ) ] ⟩ follows that basis in time.) What survives depends on j a only, while (a) is the retarded term of (20). The kernel D R is supported on the light cone and is infrared finite. After the renormalization described in Section 2, its action on a classical worldline is the Abraham–Lorentz–Dirac radiation reaction. This is where the soft virtual-photon contribution of the O ( q a ) term resides, and it is finite. (b) The Hadamard term. At any fixed frequency ω ≠ 0 , the on-shell phase space in (21) gives a finite N ( ω ) ∝ 1 / | ω | , which is smooth as the regulator is removed. A logarithm appears only in quantities integrated down to zero frequency, such as the variance (13) or the distinguishability exponent (12). The lower limit is then set by the finite duration of the observation, ω IR ∼ T obs − 1 . This is precisely the averaging over unresolved soft modes implicit in any measurement of finite time and frequency resolution. The itinerant-vacuum description, therefore, contains the BN/KLN summation and is infrared finite at every finite T obs . The 1 / f spectrum is a finite frequency dependence of S J ( ω ) , not an uncanceled divergence. A time-dependent growth of the entanglement between charged matter and low-frequency photons for dressed states has been found independently in Ref. [69].
Fukuyama’s source-free conformal argument [17] remains an important boundary condition. A free Maxwell field in four-dimensional de Sitter space has no secular infrared amplification and, hence, provides only a dry fluctuation. The QED application in Section 4, by contrast, concerns a soft field continuously driven by a massive, scattering current. The distinction needed here is, therefore, not “BN/KLN versus noise” but fixed asymptotic dressing versus an evolving dressing history. Appendix F gives the corresponding density-matrix and soft-overlap formulation.

6. Soft Gravitons, Metric Noise, and Gravitational Memory

6.1. Gravitational Influence Functional

Gravity has an infrared structure parallel to QED but one that is governed by energy-momentum rather than charge. We construct it in analogy with Section 4. The parallel is photon field A μ → graviton field h μ ν (the metric perturbation around a background g μ ν ), conserved current j μ → stress tensor T μ ν , and the Ward–Takahashi identity → diffeomorphism (stress-energy conservation).
Just as we projected the photon Hadamard kernel D μ ν H onto the current, we project the graviton Hadamard kernel onto the stress tensor. The graviton field has the Hadamard (symmetric) two-point function
D μ ν α β H ( x − x ′ ) = ⟨ { h μ ν ( x ) , h α β ( x ′ ) } ⟩ = ∫ d 4 k ( 2 π ) 4 2 π δ ( k 2 ) Λ μ ν α β ( k ) e − i k ( x − x ′ ) ,
with Λ μ ν α β being the transverse-traceless projector defined in (A9). This is the spin-2 counterpart of the photon Hadamard kernel (21); like it, it is a regular distribution and carries no infrared divergence by itself. The metric perturbation h μ ν couples to the stress tensor through 1 2 ∫ h μ ν T μ ν , so separating the soft graviton sector and projecting (42) onto the stress-energy source gives a closed-time-path influence functional for h μ ν ,
Γ g [ h r , h a ] = ∫ d 4 x h a μ ν G μ ν ( 1 ) [ h r ] − 8 π G ⟨ T ^ μ ν ⟩ + ∫ d 4 x d 4 x ′ h a μ ν ( x ) H μ ν α β ( x , x ′ ) h r α β ( x ′ ) + i 2 ∫ d 4 x d 4 x ′ h a μ ν ( x ) N μ ν α β ( x , x ′ ) h a α β ( x ′ ) + ⋯ ,
whose imaginary part is the stress-tensor noise kernel
N μ ν α β ( x , x ′ ) = 1 2 ⟨ { t ^ μ ν ( x ) , t ^ α β ( x ′ ) } ⟩ , t ^ μ ν = T ^ μ ν − ⟨ T ^ μ ν ⟩ .
The factor 1 2 in (44) follows the standard stochastic-gravity normalization [27], whereas the photon and graviton Hadamard functions (21) and (42) were defined without it; the difference is a bookkeeping convention that cancels in every observable spectrum, which depends only on the product of kernel and response. This is the source-projected form of (42), just as N QED μ ν was the current-projected form of D μ ν H ; the infrared divergence appears only upon this projection and integration over the soft-graviton phase space, not in (42) itself. It is positive by construction, and diffeomorphism invariance requires
∇ μ N μ ν α β ( x , x ′ ) = 0 ,
the gravitational analogue of the Ward–Takahashi identity (23). As in the QED case, (43) is diagrammatically a single graviton line exchanged between two stress-tensor insertions: H μ ν α β is this diagram uncut, and N μ ν α β is the same diagram with the internal graviton line cut on shell, 2 π δ ( k 2 ) Λ μ ν α β , the direct gravitational analogue of the real-time cutting rule discussed in Section 2. Stress-energy conservation (45) plays the role that the Ward–Takahashi identity plays in QED, guaranteeing that this cut is transverse and, hence, physical; it is the diagrammatic underpinning of the universality of Weinberg’s soft-graviton theorem, just as (23) underlies the universality of the soft-photon factor.
The same influence functional and its Einstein–Langevin consequence below were obtained by Hu and Verdaguer in the stochastic-gravity program [26,27] by treating the quantum matter fields as an environment. As explained in Section 2, once the reduced effective action for the metric perturbation is defined, no further coarse graining is required: the noise kernel (44) is the soft-graviton Hadamard kernel projected onto the stress-energy source, and its Hubbard–Stratonovich separation is an identity of the generating functional. The unstable gravitational infrared sector self-selects the metric perturbation as the order parameter.

6.2. Einstein–Langevin Equation

Applying the Hubbard–Stratonovich identity to (43) yields a c-number stochastic metric source ξ μ ν with ⟨ ξ μ ν ξ α β ⟩ = N μ ν α β , and the Einstein–Langevin equation [27],
G μ ν ( 1 ) [ h ] − 8 π G ⟨ T ^ μ ν ⟩ + ∫ d 4 x ′ H μ ν α β ( x , x ′ ) h α β ( x ′ ) = 8 π G ξ μ ν ( x ) .
This is not a phenomenological addition to semiclassical gravity: it is the CTP-derived representation obtained by separating the soft graviton sector, in complete parallel with the QED Langevin equation (28).

6.3. Soft Gravitons, 1 / f Metric Noise, and Memory

From high-energy scattering, the Weinberg soft theorem [38,70] gives the leading soft-graviton emission amplitude as
S g ( k ) = κ ∑ i η i p i μ p i ν ϵ μ ν ( k ) p i · k , κ 2 = 32 π G ,
which exponentiates in inclusive quantities. On the CTP contour the exponentiated object is the stress-tensor influence functional (43). In the soft limit the noise kernel behaves as
N μ ν α β soft ( ω ) ∝ G E 2 f g ( β ) | ω | Λ μ ν α β ( n ^ ) ,
the gravitational analogue of (26): the charge e 2 is replaced by G E 2 (the gravitational “charge” is energy), the transverse projector by the transverse-traceless one, and the dipole velocity factor f ( β ) ∼ β 2 by the quadrupole one f g ( β ) ∼ β 4 . The derivation parallels Appendix C and is given in Appendix D.
Gravitational memory as the retarded kernel. The gravitational memory effect [71,72]—the permanent displacement of test masses after a burst of radiation—is encoded in the retarded kernel H μ ν α β evaluated at ω → 0 . In the CTP language, memory and stochastic metric fluctuations are the causal and symmetric parts of the same influence functional. They, therefore, probe the same soft sector but need not have identical infrared singularity structure: the symmetric part controls metric noise/decoherence, while the retarded zero-frequency response records the memory. In particular, “memory” does not by itself mean that the retarded kernel is infrared divergent. This is consistent with the infrared triangle [38]: Strominger and Zhiboedov [39] established that gravitational memory is a transition between degenerate BMS vacua and that its Fourier transform reproduces the soft-graviton theorem. In the present language, the ω → 0 retarded kernel is the memory (the vacuum shift), while the symmetric kernel at finite ω is the noise (the uncertainty in the shift); the soft theorem relates the two.
Dressed gravitational scattering states have overlaps [28,29]
D i j grav ∼ exp − 1 2 B i j grav ln ω max ω IR + i Φ i j grav ,
where the imaginary phase encodes gravitational memory and the real exponent gives reduced-state decoherence. The degree of this decoherence is generically partial and depends on the scattering kinematics and on the resolution scale [30,31], exactly as the finite-observation-time logarithm here keeps D i j finite until the strict infrared limit is taken. The Anderson-orthogonality-like catastrophe of the gravitational soft sector is, therefore, the gravitational counterpart of the QED soft-sector decoherence discussed in Appendix F.

7. Non-Equilibrium Phase Transitions and Order-Parameter Dynamics

7.1. Setting

A phase transition is the clearest realization of the self-selection principle of Section 2: the order parameter does not exist as a classical variable before the transition, yet emerges as the dominant dynamical entity after it, with the remaining modes acting as its effective environment. Following the original in–in treatment of [20,22], the dissipation and noise are induced by the radiative corrections of the self-interacting unstable field itself, not by tracing out an external bath (The unstable field was one of the three systems for which a Langevin equation was first derived from the CTP effective action in [20], alongside the thermal field and cosmological particle creation; the phase-transition case was developed in [22]).
Consider an order-parameter field Φ with an unstable (double-well) potential U ( Φ ) ,
S [ Φ ] = ∫ d 4 x 1 2 ( ∂ Φ ) 2 − U ( Φ ) , U ( Φ ) = − 1 2 μ 2 Φ 2 + λ 4 ! Φ 4 ,
symmetric at high temperature and unstable (spinodal) below the critical temperature. In parallel with the de Sitter scalar in Section 8, the self-interaction λ Φ 4 generates, through its own in–in radiative corrections, both a retarded memory kernel and a positive noise kernel; the long-wavelength (infrared) modes of Φ that go unstable are the source of the divergence, and the retarded and symmetric kernels are two projections of the same self-energy (Additional microscopic fields χ coupled to Φ can be included; they contribute radiative corrections of the same structure but are not required for the mechanism).

7.2. CTP Effective Action and Noise Kernel

The infrared part of the effective action has the standard CTP form
Γ PT [ Φ r , Φ a ] = ∫ d 4 x Φ a □ Φ r + U eff ′ ( Φ r ) + ∫ d 4 x d 4 x ′ Φ a ( x ) Σ PT R ( x , x ′ ) Φ r ( x ′ ) + i 2 ∫ d 4 x d 4 x ′ Φ a ( x ) N PT ( x , x ′ ) Φ a ( x ′ ) .
The noise kernel is generated by the self-interaction of the order-parameter field, with the same two-loop (sunset) structure as the de Sitter kernel (55) [22,73],
N PT ( x , x ′ ) ∼ λ 2 6 [ G IR H ( x , x ′ ) ] 3 + ⋯ ,
G IR H being the Hadamard function of the infrared modes. The retarded kernel Σ PT R is generated by the same λ Φ 4 interaction; both are projections of a single self-energy. In the cutting-rule language of Section 2, (52) is the simultaneous cut of all three internal propagators of the two-loop sunset self-energy diagram generated by the quartic vertex, the multi-particle generalization of the single on-shell cut that produced N QED μ ν and N μ ν α β in Section 4 and Section 6: one cut line per power of the coupling that connects the two vertices, three here in place of one. In the broken phase, where Φ acquires a mean value v, expanding U ( Φ ) around it, Φ = v + ϕ , generates an effective cubic vertex λ v 6 ϕ 3 for the fluctuation ϕ , and two such vertices form a one-loop bubble with two internal lines (one external leg of each cubic vertex being taken by Φ a ). Its double cut gives an additional noise contribution
N PT mult ( x , x ′ ; Φ r ) ∼ λ 2 Φ r ( x ) Φ r ( x ′ ) [ G IR H ( x , x ′ ) ] 2 ,
which is multiplicative—its amplitude follows the order parameter—and, being of lower loop order, typically dominates over the Φ r -independent sunset term (52) once v ≠ 0 . The two contributions coexist: the additive sunset noise persists through the transition, while the multiplicative bubble noise switches on with symmetry breaking, N PT → N PT ( x , x ′ ; Φ r ) .

7.3. Order-Parameter Langevin Equation

□ Φ ( x ) + U eff ′ ( Φ ( x ) ) + ∫ d 4 x ′ Σ PT R ( x , x ′ ) Φ ( x ′ ) = − ξ PT ( x ) ,
with ⟨ ξ PT ( x ) ξ PT ( x ′ ) ⟩ = N PT ( x , x ′ ) .
Several features are notable. The noise is colored (non-Markovian), and multiplicative in the broken phase; white additive noise is, at best, an approximation near the critical point. The noise and dissipation kernels are generated by the same self-interaction, connected by the non-equilibrium fluctuation–dissipation relation. Near the critical point, Goldstone and critical modes contribute infrared-enhanced noise N PT ( ω ) ∝ 1 / | ω | α , a 1 / f -like spectrum whose exponent depends on the universality class [74,75].
Macroscopic classicality is not imposed: it emerges when the unstable infrared sector is separated from the deterministic evolution and drives the system to a particular broken-symmetry branch. The same logic applies to cosmic inflation as a dynamical phase transition [24].
Spinodal growth and spatial vacuum itinerancy. In the Z 2 model (50), modes with k 2 + U ′ ′ ( Φ ) < 0 grow exponentially (spinodal decomposition). For a continuous-symmetry Mexican hat U ( Φ ) = λ ( | Φ | 2 − v 2 ) 2 , the radial and transverse curvatures are m L 2 = 4 λ ( 3 r 2 − v 2 ) and m T 2 = 4 λ ( r 2 − v 2 ) . Both are negative near the origin; the transverse curvature stays negative throughout the inner slope r < v . These backgrounds are not stable vacua, yet normalizable quantum states evolve unitarily in the underlying theory, and exponentially growing mode functions can satisfy canonical Wronskian normalization. No stationary oscillator vacuum exists for those modes. Hubbard–Stratonovich separation of the positive noise kernel does not remove the deterministic instability.
Spinodal amplification can select different phases θ ( x , t ) in different regions. Their interaction, phase relaxation, coarsening, and defect motion provide spatiotemporal itinerancy of local vacuum orientations without making each domain a distinct global vacuum or Hilbert-space sector. At r = v , the radial mode is stable and the angular mode is massless. With local damping and short-range noise, diffusion of the uniform phase scales as D θ 0 ∝ V − 1 , whereas nonuniform Goldstone modes and evolving domains remain dynamical even in the thermodynamic limit. Unlike QED’s nonlocal, Gauss-law-constrained dressing, this is local-ordering dynamics; their commonality lies in infrared-sensitive state distinguishability and CTP response/noise structure, not identical dynamics or Hilbert-space sectors.

8. Self-Interacting Scalar Field in De Sitter Space (Summary)

The de Sitter case has been worked out in full detail in [24], where the λ ϕ 4 CTP effective action is derived at two-loop order and the Harrison–Zel’dovich spectrum n s ≃ 1 is reproduced. The CTP treatment of de Sitter density fluctuations goes back to [21]; we summarize the essential structure here for completeness.
A light scalar in de Sitter space with V ( ϕ ) = λ 4 ! ϕ 4 + 1 2 m 2 ϕ 2 develops an infrared-enhanced long-wavelength sector. The free-theory horizon-crossing noise [7,8] is dry [24]: it lacks a retarded partner generated by the effective action, and, therefore, carries no genuine dissipation.
With self-interaction, the two-loop CTP effective action generates both a retarded kernel Σ R and the noise kernel (see Appendix G)
N λ ( x , x ′ ) ∼ λ 2 6 a 3 ( t ) a 3 ( t ′ ) G H IR ( x , x ′ ) 3 ,
from the same sunset diagram. For the massless minimally coupled free field, the leading k − 3 infrared enhancement resides in the Hadamard component, while the free retarded propagator is infrared finite. The interaction generates a retarded self-energy as well, but this does not require it to share the leading infrared singularity of N λ . The Langevin equation
∂ t 2 + 3 H ∂ t − a − 2 ∇ 2 ϕ + V eff ′ ( ϕ ) + ∫ d 4 x ′ Σ R ( x , x ′ ) ϕ ( x ′ ) = − ξ λ ( x )
is the non-dry stochastic dynamics. In momentum space N λ ∝ k − 3 in the infrared, reproducing the Harrison–Zel’dovich spectrum. The key distinction from standard stochastic inflation is that the spectrum is dynamical—generated by interaction—rather than kinematical.
The curvature perturbation amplitude [76,77]
Δ ζ 2 ≃ H 2 8 π 2 ϵ M Pl 2
is the cosmological analogue of the current noise coefficient A J : both are the observable coefficients of an infrared logarithm, and both are determined by the microscopic parameters of the source.
Beyond the Gaussian leading order, the λ ϕ 4 interaction makes the Gaussian noise kernel field dependent at loop level, so that the noise is multiplicative—the noise amplitude depends on the field value ϕ r . The genuinely non-Gaussian vertices ( ϕ r ϕ a 3 and above) give perturbative corrections (Section 9). The resulting non-Gaussian tails in the curvature perturbation distribution are discussed in Section 9, where their connection to primordial black hole formation is noted.

9. Higher Loops, Resummation, and Non-Gaussian Noise

The rule for higher loops is simple: do not put an IRD back into a deterministic action. Construct the CTP effective action, separate its causal and symmetric parts, and represent the symmetric part stochastically.
At quadratic order in q a , the result is Gaussian noise. Higher loops produce higher powers of q a ,
Γ ng [ q r , q a ] = ∑ n ≥ 3 i n − 1 n ! ∫ d x 1 ⋯ d x n C n ( x 1 , … , x n ; q r ) q a ( x 1 ) ⋯ q a ( x n ) ,
which would correspond formally to non-Gaussian stochastic cumulants ⟨ ξ ( x 1 ) ⋯ ξ ( x n ) ⟩ c = C n .
Generalized characteristic functional and positivity. The precise statement is as follows. Collecting all pure- q a dependence of the weight into Φ [ q a ; q r ] ≡ exp { − 1 2 ∫ q a N q a + i Γ ng } , one can always write
Φ [ q a ; q r ] = ∫ D ξ P [ ξ ; q r ] e i ∫ q a ξ ,
with P being the functional Fourier transform of Φ . This is the non-Gaussian generalization of (7). P is a genuine probability functional if and only if Φ is positive definite (Bochner). This is a condition on the whole series, not on the individual C n . By Marcinkiewicz’s theorem [78], a cumulant series truncated at any finite order n ≥ 3 is never positive definite. The higher cumulants must, therefore, be handled in one of two ways. Either they are resummed into a positive structure—for instance, the compound-Poisson form exp { ∫ ν ( e i q a Δ − 1 ) } generated by discrete, independent scattering events, which is positive by construction—or, in the regime where they are suppressed (Section 2), they are treated as perturbative corrections to the correlators computed from the Gaussian Langevin dynamics. In the second case, they are not reinterpreted as an independent noise.
In QED, leading soft divergences exponentiate [2,5,6,70]. Higher-loop soft divergences should be organized as corrections to the retarded current–current kernel, the symmetric noise kernel (26), possible higher-current cumulants, and finite-matching coefficients. This avoids double counting of self-energy and vertex contributions.
In gravity, higher loops must respect the transversality constraint (45) at every order. Near-critical phases and gravitational collapse can produce strongly non-Gaussian metric fluctuations requiring the full cumulant expansion.
Multiplicative noise and non-Gaussianity in de Sitter  λ ϕ 4  theory. The de Sitter case with λ ϕ 4 interaction deserves special attention beyond the leading two-loop result of [24]. The quartic vertex couples four powers of ϕ , so in Keldysh variables the interaction generates terms of the form ϕ r 3 ϕ a (classical drift) and ϕ r ϕ a 3 (beyond-Gaussian CTP term). The latter is cubic in ϕ a , so it contributes to Γ ng (58) with n = 3 . Formally, it is a local, field-dependent third cumulant C 3 ∝ λ ϕ r , whose Fourier transform (an Airy-type kernel) is not positive. Restoring ℏ, it is suppressed relative to the drift vertex by ℏ 2 , or by ( ρ / F ) 2 in the infrared-enhanced regime. It is precisely the term dropped in the classical-statistical approximation [50], and we treat it perturbatively, as described above. The multiplicative character of the noise has a different origin. At loop level the Gaussian kernel itself depends on ϕ r : the one-loop bubble from Section 7 gives N ∝ λ 2 ϕ r ( x ) ϕ r ( x ′ ) [ G IR H ( x , x ′ ) ] 2 [cf. (53)], so the noise strength depends on the field value ϕ r —the noise is multiplicative while each increment remains Gaussian.
A multiplicative stochastic Langevin equation for the order parameter ϕ ,
ϕ ˙ = − V eff ′ ( ϕ ) 3 H + η 0 ( t ) + η λ ( t ) g ( ϕ ) , ⟨ η λ ( t ) η λ ( t ′ ) ⟩ = N λ δ ( t − t ′ ) ,
with g ( ϕ ) ∝ ϕ , differs fundamentally from an additive Gaussian process. The probability distribution P ( ϕ , t ) satisfies a Fokker–Planck equation with field-dependent diffusion coefficient, which generically produces non-Gaussian tails—specifically, the distribution decays more slowly than Gaussian for large | ϕ | , often approaching a power-law or exponential-tail form.
Connection to primordial black holes. This non-Gaussian tail has direct observational consequences. The abundance of primordial black holes (PBHs) is exponentially sensitive to the tail of the curvature perturbation distribution P ( R ) : even a modest enhancement of the tail beyond the Gaussian prediction can increase the PBH abundance by many orders of magnitude. Recent work on stochastic inflation in the ultra-slow-roll regime [32,33,79] has demonstrated that non-Gaussian exponential tails are a generic consequence of quantum diffusion in nonlinear potentials, and that they dramatically enhance PBH formation compared to Gaussian estimates.
In the present CTP framework, the mechanism is transparent. The λ ϕ 4 interaction generates a noise kernel N λ ( x , x ′ ) (55) and, at loop level, a field-dependent (multiplicative) part of the same Gaussian kernel. Through the Fokker–Planck equation, field-dependent diffusion together with the nonlinear drift produces a distribution of ϕ with heavier tails than a Gaussian of the same variance. This is the same mechanism that generates exponential tails in stochastic inflation [32,33]. The genuinely quantum cumulants C n ≥ 3 (e.g., from ϕ r ϕ a 3 ) only add small corrections to this tail. Combined with the long-memory structure of the 1 / f -like infrared kernel, the rare large- ϕ events are further enhanced compared to a Markovian Gaussian estimate. The CTP effective-action framework, therefore, identifies a direct path from the λ ϕ 4 IRD to non-Gaussian tails that could source PBH overproduction [34,80].
The systematic computation of the field-dependent noise kernel and of the perturbative corrections from  C 3 and C 4 from the two-loop and higher CTP effective action for λ ϕ 4 de Sitter scalar theory, and the resulting prediction for the PBH mass spectrum, is a natural extension of the framework developed here and of the results in [24].

10. Observational Consequences and Falsifiability

The measurable quantity is not the divergence itself, but the coefficient of the induced classical observable S obs ( ω ) = | χ ( ω ) | 2 N IR ( ω ) . Table 1 summarizes the four arenas.
Table 1. Infrared sectors and their observable classical signatures in the CTP framework. The many-body row is discussed in Appendix F.
QED conductor (primary prediction). The key falsifiable prediction is the impurity-density dependence (35). In the dilute-impurity regime with C IR ∝ n imp s and γ imp ∝ n imp ,
A J ( n imp ) ∝ n imp s − 2 .
The minimal soft-dressing estimate ( s = 1 , one soft-dressing event per scattering) gives A J ∝ n imp − 1 . A direct experimental protocol would use a controlled family of samples with varying impurity density while controlling carrier density, geometry, and temperature. The low-frequency noise should be fitted as
S I ( f ) = A I f α + S I white + S I other ( f ) ,
with the infrared QED contribution expected to have α ≃ 1 . Additional checks include (i) a finite bandwidth over which α ≃ 1 ; (ii) systematic dependence of A I on impurity density and mobility; (iii) logarithmic bandwidth dependence ⟨ δ I 2 ⟩ IR ≃ A I ln ( f max / f min ) ; (iv) independence from temperature at T → 0 (distinguishing from Johnson–Nyquist noise [66,67]); and (v) the absence of an intrinsic low-frequency cutoff: the flattening point of the spectrum should track ω IR ∼ T obs − 1 as the observation time is extended, without saturating. Observation of a T obs -independent flattening frequency ω c would falsify the soft-photon mechanism, whereas superpositions of Lorentzians with a bounded relaxation-time distribution necessarily produce one. The robustness of the 1 / f signal under propagation and nonlinear (thresholded) readout, relevant to its appearance across many downstream systems, has been examined in the wave-packet model [16].
Gravitational observables. The soft-graviton contribution (48) implies metric noise S h ( f ) ∝ G E 2 / f . It also predicts a permanent memory displacement Δ h mem , which is detectable in principle by pulsar timing arrays and future space-based interferometers [71,72].

11. Relation to Coarse Graining, Decoherence, and Prior Work

Coarse graining. The standard route [26,27] chooses a system–environment split and traces over the environment. The present construction does not deny this reduction: once an effective action Γ [ q ] for the collective variable is written, the unresolved degrees of freedom have already been integrated out. The self-selection principle in Section 2 specifies how that collective variable may emerge dynamically. What requires no further coarse graining is the Hubbard–Stratonovich representation of the positive symmetric kernel already present in Γ [ q ] .
Decoherence. The imaginary influence action suppresses CTP histories with large q a , favoring a classical stochastic description. No separate decoherence postulate is needed—but a genuine system–environment (or self-selected order-parameter/effective-environment) split is needed, and this is the operative distinction with respect to Ref. [81], which shows that a closed, unitarily evolving quantum system with no bipartition into system and environment cannot decohere intrinsically from squeezing alone. That result is correct and is not in tension with the present mechanism: it is the statement that applies before any order-parameter/ effective-environment split has occurred. The self-selection principle in Section 2 is precisely what supplies such a split dynamically, without it being imposed by hand; decoherence and the Langevin noise N then arise together, from the same imaginary part of Γ , once the split exists. Neither can appear without it, which is why Ref. [81], working in a single global Hilbert space with no such split, correctly finds neither. Whether the stochastic equation becomes a macroscopic record, an order parameter, or a cosmological perturbation is a subsequent physical question. In the itinerant-vacuum language, decoherence is the statement that different classical histories—corresponding to different vacuum trajectories—cannot interfere with: the orthogonality of the dressed states is the same phenomenon as decoherence, viewed from the Hilbert-space rather than the density-matrix perspective. This identification is quantitatively the same one that underlies horizon-induced decoherence [41] and infrared quantum information [28,29,30,31], where the soft which-path memory plays the role of the environment record.
Asymptotic dressing and real-time itinerancy. For QED it is economical to regard the FK coherent state as the asymptotic limit of the present construction. A fixed scattering process is characterized by its in/out dressing, whereas the real-time CTP description resolves the interpolation α in → α [ J ( t ) ] → α out . BN/KLN and FK ensure the usual infrared-safe scattering physics; the CTP/Hubbard–Stratonovich representation asks how the same soft sector appears when the collective source itself evolves. This viewpoint keeps the standard QED result intact while exposing memory, decoherence, and noise that are invisible in a single in–out label.
Checks required in concrete applications.
(i)
Use the in–in CTP effective action, not the in–out action.
(ii)
Decompose the IRD-bearing kernel into a causal retarded part and a positive symmetric part before varying the action.
(iii)
Respect Ward identities (23), diffeomorphism transversality (45), and internal-symmetry constraints on the noise kernel.
(iv)
Verify the two conditions for genuine classical stochastic dynamics. (a) Do not use dry noise: any stochastic source must have an interaction-generated retarded partner. A conformally protected free field (e.g., source-free Maxwell in de Sitter) is dry and does not classicalize, in agreement with [17]; the interacting sector coupled to a charged current is not. (b) Verify anticommutator dominance where a mode occupation is meaningful, and for coherent soft dressings also verify cumulative overlap suppression between distinct histories. A divergent total soft occupation need not mean macroscopic occupation of each mode; the infrared classicality criterion is collective rather than mode-by-mode.
(v)
For soft QED, verify first that the asymptotic sector is the standard infrared-safe dressed one. The real-time problem then concerns the evolution of its dressing profile α [ J ( t ) ] , not a revival of an uncanceled inclusive divergence (Section 5).
(vi)
For 1 / f behavior, ω IR ∼ T obs − 1 ; ω max − 1 is the shortest physical time in the infrared scaling regime.

12. Conclusions

Infrared divergences are not merely pathologies of perturbation theory. They are signatures of an itinerant vacuum: a quantum vacuum that wanders continuously through a family of inequivalent states as the classical order parameter evolves under the stochastic force generated by the CTP effective action. When properly relocated from the deterministic part of the action into the stochastic part via the Hubbard–Stratonovich identity, an IRD becomes the origin of observable classical structure. The relocation is an algebraic identity; no coarse graining beyond the reduction to the collective-variable effective action, and no separate decoherence postulate, is required. The “vacuum” that wanders is the infrared-dressed sector accompanying the collective variable; the Poincaré-invariant ground state is unchanged. In QED the Bloch–Nordsieck/KLN summation is contained in the CTP trace, and the real-time result is infrared finite at every finite observation time (Section 5).
The four arenas of this paper illustrate different facets of the mechanism.
  • Soft QED and  1 / f  current noise. The soft-photon phase space generates a 1 / f  eikonal soft-history weight, Equation (26), which, under the transport matching assumptions in Section 4.3, yields A J ∝ n e 2 C IR / γ imp 2 —a conditional scaling prediction in terms of carrier density, charge, velocity, and the impurity scattering rate. This is a field-theoretically motivated implementation of Handel’s intuition [9,10], now formulated as a CTP identity that resolves the formal objections of [12,13,14]. The standard FK-coherent dressing is retained as the infrared-safe asymptotic description; the itinerant-vacuum extension promotes its fixed profile to the real-time stochastic trajectory α [ J ( t ) ] . Soft-sector orthogonality then becomes reduced-state decoherence rather than a contradiction (Section 5).
  • Soft gravitons and gravitational memory. The stress-tensor Hadamard kernel produces 1 / f metric noise with amplitude ∝ G E 2 / | ω | . Gravitational memory is encoded in the retarded zero-frequency response. Noise and memory are complementary projections of the same soft-sector influence functional, although their infrared singularity structure need not be identical.
  • Non-equilibrium phase transitions. The infrared instability of the order parameter generates a multiplicative colored-noise Langevin equation. Macroscopic classicality arises from the CTP separation, not from externally imposed decoherence.
  • De Sitter scalar fields and primordial black holes. The two-loop λ ϕ 4 noise kernel reproduces the Harrison–Zel’dovich spectrum [24]. The free-field kinematical noise is dry; the interaction-generated noise is the genuine dynamical source. Beyond the Gaussian leading order, the quartic interaction makes the noise kernel field dependent (multiplicative noise), producing non-Gaussian tails in the probability distribution of the curvature perturbation. These tails could source primordial black hole overproduction [32,33,34], connecting the CTP infrared mechanism directly to one of the most actively studied topics in current cosmology.
Higher-loop IRDs follow the same rule: organize into retarded kernels and Gaussian or non-Gaussian noise cumulants. The broader message, summarized in
infrared divergence ⟶ itinerant vacuum ⟶ classical stochastic degrees of freedom ,
connects QED soft photons, de Sitter fluctuations, gravitational memory, and critical phenomena within a common CTP framework.
In every case, the infrared divergence is not merely a defect of the theory but a sign that the vacuum is changing its state. The CTP effective action reads this sign correctly: it separates the divergence into a retarded memory of past transitions and a stochastic force toward future ones. The divergence is thereby converted from an embarrassment into an instrument of prediction.
The two ingredients of this construction—a Langevin equation for the mean field read off directly from the CTP effective action, with non-Gaussian and colored noise, and the separation of an infrared-divergent kernel by the identity alone—were both present, the first explicitly and the second in embryo, in the author’s earliest work on dissipative quantum systems [20,21]. The present paper returns to that origin and carries it to its natural conclusion: the infrared divergence is not a defect to be removed but the seed of an itinerant vacuum.
Outlook: geometry of the vacuum manifold. The itinerant-vacuum picture suggests a natural geometric extension that we sketch here and develop elsewhere. As the order parameter ϕ ( t ) traces a stochastic trajectory in the full configuration space M ϕ , one may formally associate admissible quantum states H [ ϕ 0 ] with each background ϕ 0 ∈ M ϕ , stable or unstable. These labels need not denote inequivalent Hilbert spaces: unstable backgrounds support normalizable evolving states, but not stationary fluctuation vacua. Schematically, this suggests a Hilbert-space bundle
⨆ ϕ 0 ∈ M ϕ H [ ϕ 0 ] ⟶ M ϕ ,
in which the dressed-state overlap D i j plays the role of a connection, encoding parallel transport between neighbouring fibres.
For a Mexican-hat potential U ( ϕ ) = λ ( | ϕ | 2 − v 2 ) 2 with complex ϕ , the base space is the entire complex ϕ -plane, M ϕ ≅ C ≅ R 2 . The familiar vacuum manifold VM ≅ S 1 of radius | ϕ | = v is a special submanifold of M ϕ : the locus of absolute minima. At the origin and on the slope, negative curvatures drive spinodal growth; normalizable quantum states evolve, but these backgrounds are not additional vacua. After different regions approach | ϕ | = v , their phases θ ( x , t ) interact and coarsen through Goldstone modes and defects, even when global phase diffusion is suppressed as V → ∞ (Section 7). This spatial itinerancy concerns evolving local vacuum orientations, not transitions among distinct Hilbert spaces assigned to individual domains. The CTP noise representation does not eliminate spinodal instability, and the bundle interpretation remains conjectural.
This fibre-bundle picture resolves several otherwise awkward problems. Unstable particles, whose states formally escape a fixed Hilbert space (necessitating Gamow vectors or Rigged Hilbert spaces), are naturally described as trajectories on M ϕ that drift away from an unstable fixed point toward a distant fibre—without invoking a non-standard inner product. Quantum measurement can be formulated as a stochastic process on the bundle: einselection [82] localises the trajectory onto a preferred fibre (a pointer state), while the procoherence mechanism of [83] drives the order parameter toward a definite broken-symmetry vacuum on S 1 , with the classical meter reading identified as the angular coordinate θ on the vacuum manifold. The CTP Langevin equation derived in this paper is then the equation of motion for the meter pointer itself.
A systematic development of this geometric framework—including the differential geometry of M ϕ , the curvature of the Hilbert-space bundle, and its applications to measurement theory and the classification of infrared sectors—is left to a forthcoming paper.
A separate cosmological question is whether the stress-energy fluctuations of ordinary matter, acting as a source through the gravitational sector of Section 6, leave an imprint on the dark sector of the late universe. Because such an imprint would track the matter that sources it, it could bear on the coincidence problem; but establishing its equation of state and magnitude requires a dedicated quantitative study and lies beyond the scope of this paper.

Funding

This research received no external funding

Data Availability Statement

No new data were created or analyzed in this study. Data sharing is not applicable to this article.

Acknowledgments

The author is grateful to Akika Nakamichi for valuable discussions and for collaboration on earlier studies that laid the foundation for the present work.

Conflicts of Interest

The author declares no conflict of interest.

Appendix A. Hubbard–Stratonovich Identity

For a real positive kernel N, the Gaussian identity is
∫ D ξ exp − 1 2 ξ N − 1 ξ + i q a ξ = N ′ exp − 1 2 q a N q a .
Completing the square,
− 1 2 ( ξ − i N q a ) N − 1 ( ξ − i N q a ) − 1 2 q a N q a .
If higher powers of q a are present, the stochastic process is non-Gaussian and is specified by its cumulant functional, provided the resulting characteristic functional is positive definite; otherwise the higher cumulants are treated perturbatively (Section 9).

Appendix B. Useful Formulae for the 1/ f Kernel

With the convention (14),
N 1 / f ( t ) = A π Ci ( ω max | t | ) − Ci ( ω IR | t | ) .
For ω IR | t | ≪ 1 , Ci ( ω IR | t | ) = γ E + ln ( ω IR | t | ) + O ( ( ω IR t ) 2 ) . For ω max | t | ≫ 1 , Ci ( ω max | t | ) is oscillatory and suppressed as 1 / ( ω max | t | ) . The memory is, therefore, logarithmic in the intermediate range t UV ≪ t ≪ T obs .

Appendix C. Soft QED as a Wilson-Line Influence Functional

For a charged particle with eikonal current  j μ ( x ) = e ∫ d τ u μ ( τ ) δ ( 4 ) ( x − z ( τ ) ) —i.e., the current sourced by a fixed, undeflected classical worldline z ( τ ) with four-velocity u μ ( τ ) , the approximation in which the recoil of the emitting charge under photon emission is neglected—the gauge-invariant form of (20) is
Γ γ = − ln P exp i e ∫ C + d z μ A μ P exp − i e ∫ C − d z μ A μ A .
Under a gauge transformation A μ → A μ + ∂ μ λ , each path-ordered exponential acquires only a boundary phase exp [ i e ( λ ( z f ) − λ ( z i ) ) ] ; since the C + and C − contours share the same endpoints z i , z f , these phases cancel between the two exponentials for any λ , at every order in e. Equation (A4) is, therefore, gauge invariant non-perturbatively, without appeal to (23) order by order: expanding in powers of e generates every attachment of a photon line to the worldline, with no distinction between what would be called a “self-energy” or a “vertex” insertion in the usual Feynman-diagram language: all such insertions are attachment points on the same line. Equation (23) is the statement, valid order by order, that the sum of self-energy and vertex cuts is gauge invariant; (A4) is the non-perturbative, all-orders object whose gauge invariance implies (23) rather than needing to be checked against it. Cutting (A4) on shell—replacing the exchanged photon line(s) by their Wightman functions, as in the general discussion in Section 2—therefore, gives the gauge-invariant D μ ν H of (21) directly, with no separate self-energy/vertex bookkeeping.
Expanding (A4) to the lowest order in e attaches a single soft photon of momentum k to the worldline; the near-on-shell charged propagator pole combines with the vertex to give the eikonal factor
e u μ u · k ,
universal (spin- and process-independent) because it depends only on the charge and velocity of the emitting line, not on its internal structure—the field-theoretic statement of Low’s theorem [53]. It is this factor, squared and summed over sources, that produces the denominators ( u i · k ) ( u j · k ) below; the angular dependence on x = cos θ in f ( β ) is a consequence of (A5), not its definition. For N = n V isotropic sources with speed β , the current-projected noise kernel in the soft limit is
N QED μ ν ( k ) = ∑ i , j e 2 2 π δ ( k 2 ) u i μ u j ν − u i · u j P μ ν T ( u i · k / | k | ) ( u j · k / | k | ) .
After the angular integral and particle sum,
N QED μ ν ( ω ) = n e 2 f ( β ) | ω | P μ ν T ,
with the velocity factor arising from the transverse-projected eikonal integral
f ( β ) = 1 2 ∫ − 1 1 d x β 2 ( 1 − x 2 ) ( 1 − β x ) 2 = 1 β ln 1 + β 1 − β − 2 → β → 0 2 3 β 2 ,
which is (25). The 1 / | ω | follows from the radial measure ω d ω times the eikonal 1 / ( p · k ) 2 ∝ 1 / ω 2 . This is Equation (26).
The dressing as a current-driven coherent state. It is worth noting why the Kulish–Faddeev dressing [5,6] appears automatically in this picture. A classical c-number current j μ coupled linearly to the free photon field drives each field mode like a forced oscillator, whose ground state is a Glauber coherent state with displacement α k ∼ j ˜ ( k ) / 2 ω k , i.e., an infinite superposition of soft photons. For the eikonal current j μ = e ∫ d τ u μ δ ( 4 ) ( x − z ) , this displacement is α k ∼ e u · ϵ / ( u · k ) , which is identical to the Kulish–Faddeev soft factor. The dressed asymptotic state, introduced in the S-matrix formulation as a prescription to restore infrared finiteness is, therefore, nothing but the coherent state that a classical charged current necessarily produces. The same background-current (Wilson line) model has been analyzed by Hirai and Sugishita [36], who showed that it shares the full infrared structure of QED and used it to demonstrate that the original Kulish–Faddeev dressing alone does not render the S-matrix infrared finite: the correct, gauge-invariant dress code is fixed by the asymptotic symmetry [35,37]. That refinement concerns the infrared finiteness of S-matrix elements, which the present construction does not invoke; the noise-kernel derivation above uses the coherent state only as the field configuration driven by a classical current, and is insensitive to the choice among infrared-equivalent dressings. Its mean photon number ⟨ N γ ⟩ ∼ e 2 ∫ d ω f ( β ) / ω is the same infrared logarithm that appears in the noise kernel above. In the itinerant-vacuum language, the order parameter (the current) labels a coherent vacuum; as it evolves it drives a succession of such vacua, and averaging over the unresolved soft history of this succession yields the stochastic kernel. The single dressing of a fixed scattering event and the noise of an evolving order parameter are the same coherent-state physics viewed statically and dynamically.

Appendix D. Soft Gravitons and Stress-Tensor Noise

The gravitational kernel follows the same steps as Appendix C with the replacements spin-1 → spin-2, charge → energy-momentum, and P μ ν T → Λ μ ν α β . The soft-graviton factor carries two powers of momentum, κ p μ p ν / ( p · k ) with κ 2 = 32 π G , because the source is the stress tensor T μ ν = ∑ i m i ∫ d τ i u i μ u i ν δ ( 4 ) ( x − z i ) rather than the current. The transverse-traceless projector is the spin-2 analogue of P μ ν T ,
Λ μ ν α β = 1 2 P μ α T P ν β T + P μ β T P ν α T − P μ ν T P α β T .
The same on-shell phase space and diagonal-source sum give
N μ ν α β soft ( ω ) = n g ( 32 π G ) E 2 f g ( β ) | ω | Λ μ ν α β ( n ^ ) ,
the full form of (48), with energy E = γ m replacing the charge and the spin-2 velocity factor
f g ( β ) = 1 2 ∫ − 1 1 d x β 4 ( 1 − x 2 ) 2 ( 1 − β x ) 2 → β → 0 8 15 β 4 .
The quadratic β 2 of QED (dipole radiation) is replaced by the quartic β 4 (quadrupole radiation), the only structural change between the two arenas. Static mass ( β → 0 ) does not radiate: only the time-varying quadrupole contributes to the noise, consistent with the absence of monopole and dipole gravitational radiation.
Temporal versus spatial spectrum: one infrared divergence. The 1 / | ω | spectra of QED and gravity, and the 1 / k 3 Harrison–Zel’dovich spectrum of the de Sitter field (Section 8), are two faces of the same infrared divergence, related by the on-shell condition ω = | k | . A noise kernel ∝ δ ( k 2 ) / ( p · k ) 2 gives, in the temporal description, the power spectral density N ( ω ) ∝ 1 / | ω | ; in the equal-time spatial description, the three-dimensional power spectrum
P ( k ) ∼ 1 k · 1 k 2 = 1 k 3 ,
where the 1 / k is the on-shell phase space and the 1 / k 2 is the soft (eikonal) factor. The two statements are equivalent expressions of the logarithmic infrared divergence,
∫ d ω ω ⟺ ∫ k 2 d k k 3 = ∫ d k k .
A flat temporal power ( ω N = const ) and a scale-invariant spatial power ( k 3 P = const ) are the same statement. Thus, gravitational metric noise is 1 / f when observed as a time series (pulsar timing, interferometry) and Harrison–Zel’dovich, P ( k ) ∝ 1 / k 3 , when observed as a spatial pattern—the same itinerant-vacuum infrared structure as the de Sitter spectrum, reached here through soft gravitons rather than super-horizon scalar modes.

Appendix E. Scaling Estimate for the Soft-Photon Noise Coefficient CIR

For a charged-particle scattering event with momentum transfer Δ p , the soft emission probability is
d P soft = C soft α em d ω ω ,
where C soft contains angular and kinematic factors. If scattering events occur at rate Γ scatt , the infrared stochastic force strength scales as
C IR ∼ α em C soft Γ scatt ( Δ p ) 2 .
For dilute impurities, Γ scatt ≃ n imp v F σ tr ≃ γ imp . Substituting into (35),
A J ∼ α em C soft n e 2 m 2 ( Δ p ) 2 γ imp .
If Δ p and σ tr are independent of impurity density, this gives A J ∝ n imp − 1 . This is a scaling estimate. A complete microscopic test must derive the force covariance and its response from fermionic CTP dynamics in a screened, finite-temperature, disordered electron gas; the soft-emission weight alone does not establish Equation (A15).

Appendix F. Relation to Handel’s Infrared Theory of 1/f Noise

Handel proposed that 1 / f noise in electric currents is a fundamental QED phenomenon associated with soft-photon radiative corrections to charged-particle scattering [9,10]. This was an important and suggestive idea: it identified a possible connection between the universality of low-frequency current noise and the universal soft structure of QED.
Objections to Handel’s theory. The proposal was criticized by several authors [12,13,14]. The central objections are as follows:
1.
Scattering states with different soft-photon content are orthogonal in the infrared limit (soft-sector orthogonality):
⟨ γ j | γ i ⟩ soft soft → 0 as ω IR → 0 .
This makes it problematic to identify interference between different soft sectors as an ordinary current fluctuation in an S-matrix framework.
2.
A direct identification of infrared-divergent cross sections with current fluctuations is not rigorously justified.
The objections as constructive contributions. The criticisms of Refs. [12,13,14] deserve appreciation as physically precise analyses that sharpen the question rather than close it. Nieuwenhuizen et al. and Kiss–Heszler correctly identified that soft-sector orthogonality makes interference between different dressing states ill-defined in an S-matrix framework: this is not a mere technical complaint, but a genuine physical observation about the structure of QED Hilbert space. Weissman’s comprehensive review [12] further established that any proposal connecting QED infrared structure to 1 / f noise must provide a rigorous bridge between cross sections and fluctuation spectra—a bridge that Handel’s original formulation lacked.
These are not objections to be dismissed; they are structural constraints that the correct theory must satisfy. The present CTP formulation is designed to satisfy all of them. Together, Handel’s intuition and its critics collectively charted the logical territory of the problem.
The CTP resolution. In the present formulation both objections are resolved. For a physical charged state dressed by soft photons,
| Ψ ⟩ = ∑ i c i | p i ⟩ hard ⊗ | γ i ⟩ soft ,
tracing over unresolved photons gives the reduced density matrix with decoherence factors
ρ hard i j = c i c j * D i j | p i ⟩ ⟨ p j | , D i j = ⟨ γ j | γ i ⟩ soft soft = exp − 1 2 B i j ln ω max ω IR + i Φ i j .
When the soft clouds become orthogonal, D i j → 0 and interference between different hard histories is suppressed. This is reduced-state decoherence, not a failure of the description. At any finite resolution the suppression is partial: D i j remains nonzero, with a magnitude controlled by the kinematics and by the infrared resolution scale, as analyzed in detail in [30,31]; complete orthogonality holds only in the strict limit ω IR → 0 . Objection (1) is, therefore, not a contradiction but a statement that the soft sector carries which-path information. In the itinerant-vacuum language of this paper, the orthogonality D i j → 0 simply means that a charged particle that has moved from momentum p i to p j now inhabits a genuinely different vacuum (infrared-dressed sector): the dressed coherent state has shifted from | γ [ p i ] ⟩ to | γ [ p j ] ⟩ , and these are orthogonal because the soft cloud has moved. Note that the exponent B i j ln ( ω max / ω IR ) in (A19) is the same ∫ d ω / ω integral that makes ⟨ N γ ⟩ diverge: the divergence of the soft occupation and the orthogonality of the dressed vacua are the same infrared fact, viewed as a number and as an overlap. The 1 / f noise is the classical trace of this continuous vacuum wandering.
Objection (2) is resolved by replacing the direct cross-section argument with the CTP chain (1). The reduced in–in dynamics is written as a Schwinger–Keldysh influence action. The infrared part of its imaginary component is
Im Γ IF IR = 1 2 J Δ N IR J Δ .
By the Hubbard–Stratonovich identity,
exp − Im Γ IF IR = ∫ D ξ P [ ξ ] exp i ∫ ξ J Δ , ⟨ ξ ( ω ) ξ ( − ω ) ⟩ = N IR ( ω ) .
The route to 1 / f noise is (1), not a direct identification of an infrared-divergent cross section with a current fluctuation.
FK dressing as the static limit. For a fixed scattering history the current produces one coherent soft profile | α [ J ] ⟩ , which is precisely the structure incorporated in FK asymptotic states. The dressed formulation and the Bloch–Nordsieck construction give the same infrared-safe cross sections [18,19]. The real-time extension is, therefore, simply
| α in ⟩ ⟶ | α [ J ( t ) ] ⟩ ⟶ | α out ⟩ , N → ξ ( t ) → J ( t ) → α [ J ( t ) ] .
The auxiliary regulator required in an intermediate S-matrix calculation is removed in the standard infrared-safe construction. The lower scale ω IR ∼ T obs − 1 in the real-time kernel is, instead, the physical resolution of a finite observation. What we call an itinerant vacuum is this evolving family of infrared-dressed sectors, not a modification of the bare QED vacuum.
Physical analogy: Anderson orthogonality catastrophe. The soft-photon decoherence factor (A19) is structurally identical to the Anderson orthogonality catastrophe in a Fermi sea [84,85],
| ⟨ Ψ 0 ′ | Ψ 0 ⟩ | 2 ∼ exp − α AOC ln E UV E IR ,
where α AOC is set by scattering phase shifts and the infrared cloud is made of low-energy particle-hole pairs. Observable consequences include Fermi-edge singularities A ( ω ) ∼ ( ω − ω th ) − α FES and Loschmidt echoes [86,87]. In a disordered conductor, both soft photons and particle-hole pairs may contribute to low-frequency noise:
N total = N soft photon + N particle - hole + N defects + ⋯ .
The present paper focuses on the soft-photon contribution; the particle-hole term is an important complementary channel that deserves a parallel CTP treatment.
Scaling comparison with Handel. At the level of scaling, the present approach preserves Handel’s central intuition. The coefficient A J is controlled by charged-particle scattering events that update the soft electromagnetic dressing. The difference is interpretation: Handel attributed the noise to a divergent cross section, whereas here it arises from the trace over the soft sector encoded in the CTP influence action. The fine-structure-constant dependence A J ∝ α em (via C IR ∝ α em , see Appendix E) is present in both pictures, but the amplitude formula (35) also contains the impurity-density and velocity factors absent from Handel’s universal coefficient.

Appendix G. Interaction-Generated de Sitter Noise Kernel

For λ ϕ 4 theory in de Sitter space, the CTP contour interaction is
S int [ ϕ + ] − S int [ ϕ − ] = − ∫ d 4 x a 3 ( t ) λ 4 ! ( ϕ + 4 − ϕ − 4 ) .
In Keldysh variables, ϕ + 4 − ϕ − 4 = 4 ϕ r 3 ϕ a + ϕ r ϕ a 3 . The first term gives the classical drift V ′ ( ϕ r ) ; loop corrections generate retarded and noise kernels. The schematic two-loop structure
N λ ( x , x ′ ) ∝ λ 2 [ G H ( x , x ′ ) ] 3
shows that noise requires interaction: a purely kinematical, dry stochastic source cannot produce this kernel.

Appendix H. Coherent Resonance and Itinerant Vacua

A useful comparison, suggested by the referee, is with resonating quantum states. The Resonating Valence Bond (RVB) state proposed for correlated electrons is a coherent superposition of valence-bond coverings [88]; benzene resonance and ammonia inversion are finite-dimensional analogues in the broad sense of coherent superposition or tunneling between configurations [89,90]. These should not be called distinct QFT vacua. Their common diagnostic with the present problem is the overlap of the states that accompany the alternatives. Let each configuration a of the system be correlated with a state | E a ⟩ of its environment—for the present problem, the soft cloud | α a ⟩ :
| Ψ ⟩ = ∑ a c a | χ a ⟩ ⊗ | E a ⟩ .
The interference between a and b in the reduced state is then weighted by the record overlap
K a b = ⟨ E b | E a ⟩ .
When K a b remains finite, the alternatives form one coherent resonating state. For infrared-dressed histories, by contrast, K a b = ⟨ α b | α a ⟩ ∼ e − D a b can vanish as the soft sector accumulates, and the reduced dynamics becomes a stochastic itinerary among effectively orthogonal sectors. Thus, coherent resonance and itinerant vacua are related but opposite limits of the same overlap criterion: coherence binds the former, whereas infrared orthogonalization classicalizes the latter.
It is the record overlap (A28) that decides, not the overlap ⟨ χ b | χ a ⟩ of the alternatives themselves. Macroscopically different valence-bond coverings are nearly orthogonal in the thermodynamic limit, yet the RVB state stays coherent because no environment records which covering is realized. The crossover is controlled by the resonance (tunneling) amplitude Δ and the rate Γ dec at which records accumulate. For Δ ≫ Γ dec one has coherent resonance, with a level splitting. For Δ ≪ Γ dec one has incoherent, stochastic hopping among effectively orthogonal sectors—an elementary version of stochastic itinerancy. Ammonia provides an experimentally documented example. In a dilute gas the inversion doublet displays coherent resonance. With increasing pressure, collisions record the pyramidal configuration, the inversion line shifts to zero frequency, and the molecules localize in their classical configurations [91,92]. The same environment-induced localization underlies the stability of chiral molecules [93]. In the present problem the “environment” is the soft photon sector itself. Because D a b accumulates logarithmically toward the infrared, the record is always eventually written for distinct current histories. The itinerant vacuum is, in this sense, the infrared, continuum analogue of the high-pressure regime of the ammonia molecule: coherent resonance, once embedded in an infrared environment, evolves into a stochastic itinerary through orthogonal sectors.
In an RVB superconductor the microscopic resonating state can support a macroscopic superconducting order parameter (and, hence, supercurrent); this is analogous to, but microscopically distinct from, the current order parameter that labels the QED dressing trajectory here.

References

  1. Bloch, F.; Nordsieck, A. Note on the Radiation Field of the Electron. Phys. Rev. 1937, 52, 54–59. [Google Scholar] [CrossRef] [Scilit]
  2. Yennie, D.R.; Frautschi, S.C.; Suura, H. The Infrared Divergence Phenomena and High-Energy Processes. Ann. Phys. 1961, 13, 379–452. [Google Scholar] [CrossRef] [Scilit]
  3. Kinoshita, T. Mass Singularities of Feynman Amplitudes. J. Math. Phys. 1962, 3, 650–677. [Google Scholar] [CrossRef] [Scilit]
  4. Lee, T.D.; Nauenberg, M. Degenerate Systems and Mass Singularities. Phys. Rev. 1964, 133, B1549–B1562. [Google Scholar] [CrossRef] [Scilit]
  5. Kibble, T.W.B. Coherent Soft-Photon States and Infrared Divergences. IV. The Scattering Operator. Phys. Rev. 1968, 175, 1624–1640. [Google Scholar] [CrossRef] [Scilit]
  6. Kulish, P.P.; Faddeev, L.D. Asymptotic Conditions and Infrared Divergences in Quantum Electrodynamics. Theor. Math. Phys. 1970, 4, 745–757. [Google Scholar] [CrossRef] [Scilit]
  7. Starobinsky, A.A. Stochastic de Sitter (Inflationary) Stage in the Early Universe. Lect. Notes Phys. 1986, 246, 107–126. [Google Scholar] [CrossRef] [Scilit] [PubMed]
  8. Starobinsky, A.A.; Yokoyama, J. Equilibrium State of a Self-Interacting Scalar Field in the de Sitter Background. Phys. Rev. D 1994, 50, 6357–6368. [Google Scholar] [CrossRef] [Scilit] [PubMed]
  9. Handel, P.H. 1/f Noise—An “Infrared” Phenomenon. Phys. Rev. Lett. 1975, 34, 1492–1495. [Google Scholar] [CrossRef] [Scilit]
  10. Handel, P.H. Quantum Approach to Noise. Phys. Rev. A 1980, 22, 745–757. [Google Scholar] [CrossRef] [Scilit]
  11. Dutta, P.; Horn, P.M. Low-Frequency Fluctuations in Solids: 1/f Noise. Rev. Mod. Phys. 1981, 53, 497–516. [Google Scholar] [CrossRef] [Scilit]
  12. Weissman, M.B. 1/f Noise and Other Slow, Nonexponential Kinetics in Condensed Matter. Rev. Mod. Phys. 1988, 60, 537–571. [Google Scholar] [CrossRef] [Scilit]
  13. Nieuwenhuizen, T.M.; Frenkel, D.; van Kampen, N.G. Objections to Handel’s Quantum Theory of 1/f Noise. Phys. Rev. A 1987, 35, 2750–2753. [Google Scholar] [CrossRef] [Scilit] [PubMed]
  14. Kiss, L.B.; Heszler, P. An Exact Proof of the Invalidity of Handel’s Quantum 1/f Noise Model, Based on Quantum Electrodynamics. J. Phys. C Solid State Phys. 1986, 19, L631–L633. [Google Scholar] [CrossRef] [Scilit]
  15. Morikawa, M.; Nakamichi, A. A simple model for pink noise from amplitude modulations. Sci. Rep. 2023, 13, 8364. [Google Scholar] [CrossRef] [Scilit] [PubMed]
  16. Morikawa, M.; Nakamichi, A. Pink Noise in Electric Current from Amplitude Modulations. In Proceedings of the 2023 First International Conference on Advances in Electrical, Electronics and Computational Intelligence (ICAEECI), Tiruchengode, India, 19–20 October 2023; IEEE: New York, NY, USA, 2023. [Google Scholar] [CrossRef] [Scilit]
  17. Fukuyama, T. Infrared Divergence in QED and the Fluctuation of Electromagnetic Fields. arXiv 2025, arXiv:2506.08269. [Google Scholar]
  18. Choi, S.; Akhoury, R. Subleading Soft Dressings of Asymptotic States in QED and Perturbative Quantum Gravity. J. High Energy Phys. 2019, 2019, 31. [Google Scholar] [CrossRef] [Scilit]
  19. Chung, V. Infrared Divergence in Quantum Electrodynamics. Phys. Rev. 1965, 140, B1110–B1122. [Google Scholar] [CrossRef] [Scilit]
  20. Morikawa, M. Classical fluctuations in dissipative quantum systems. Phys. Rev. D 1986, 33, 3607–3612. [Google Scholar] [CrossRef] [Scilit] [PubMed]
  21. Morikawa, M. The Origin of the Density Fluctuations in de Sitter Space. Prog. Theor. Phys. 1987, 77, 1163–1177. [Google Scholar] [CrossRef] [Scilit][Green Version]
  22. Morikawa, M. Cosmological Inflation as a Quantum Phase Transition. Prog. Theor. Phys. 1995, 93, 685–709. [Google Scholar] [CrossRef][Green Version]
  23. Morikawa, M. Infrared Divergence Separated for Stochastic Force: Langevin Evolution in the Inflationary Era. arXiv 2016, arXiv:1604.01015. [Google Scholar]
  24. Morikawa, M. Quantum Fluctuations in Vacuum Energy: Cosmic Inflation as a Dynamical Phase Transition. Universe 2022, 8, 295. [Google Scholar] [CrossRef] [Scilit]
  25. Calzetta, E.; Hu, B.L. Closed-time-path functional formalism in curved spacetime: Application to cosmological back-reaction problems. Phys. Rev. D 1987, 35, 495–509. [Google Scholar] [CrossRef] [Scilit] [PubMed]
  26. Calzetta, E.A.; Hu, B.L.B. Nonequilibrium Quantum Field Theory; Cambridge University Press: Cambridge, UK, 2008. [Google Scholar] [CrossRef] [Scilit]
  27. Hu, B.L.; Verdaguer, E. Stochastic Gravity: Theory and Applications. Living Rev. Relativ. 2008, 11, 3. [Google Scholar] [CrossRef] [Scilit]
  28. Carney, D.; Chaurette, L.; Neuenfeld, D.; Semenoff, G.W. Infrared Quantum Information. Phys. Rev. Lett. 2017, 119, 180502. [Google Scholar] [CrossRef] [Scilit] [PubMed]
  29. Carney, D.; Chaurette, L.; Neuenfeld, D.; Semenoff, G.W. Dressed Infrared Quantum Information. Phys. Rev. D 2018, 97, 025007. [Google Scholar] [CrossRef] [Scilit]
  30. Gómez, C.; Letschka, R.; Zell, S. Infrared Divergences and Quantum Coherence. Eur. Phys. J. C 2018, 78, 610. [Google Scholar] [CrossRef] [Scilit] [PubMed]
  31. Gómez, C.; Letschka, R.; Zell, S. The Scales of the Infrared. J. High Energy Phys. 2018, 9, 115. [Google Scholar] [CrossRef] [Scilit]
  32. Ezquiaga, J.M.; García-Bellido, J.; Vennin, V. The Exponential Tail of Inflationary Fluctuations: Consequences for Primordial Black Holes. J. Cosmol. Astropart. Phys. 2020, 2020, 29. [Google Scholar] [CrossRef] [Scilit]
  33. Figueroa, D.G.; Raatikainen, S.; Räsänen, S.; Tomberg, E. Non-Gaussian Tail of the Curvature Perturbation in Stochastic Ultraslow-Roll Inflation: Implications for Primordial Black Hole Production. Phys. Rev. Lett. 2021, 127, 101302. [Google Scholar] [CrossRef] [Scilit] [PubMed]
  34. Bullock, J.S.; Primack, J.R. Non-Gaussian Fluctuations and Primordial Black Holes from Inflation. Phys. Rev. D 1997, 55, 7423–7439. [Google Scholar] [CrossRef] [Scilit]
  35. Hirai, H.; Sugishita, S. Dressed States from Gauge Invariance. J. High Energy Phys. 2019, 6, 23. [Google Scholar] [CrossRef] [Scilit]
  36. Hirai, H.; Sugishita, S. IR Finite S-Matrix by Gauge Invariant Dressed States. J. High Energy Phys. 2021, 2, 25. [Google Scholar] [CrossRef] [Scilit]
  37. Hirai, H.; Sugishita, S. Dress Code for Infrared Safe Scattering in QED. Prog. Theor. Exp. Phys. 2023, 2023, 053B04. [Google Scholar] [CrossRef] [Scilit]
  38. Strominger, A. Lectures on the Infrared Structure of Gravity and Gauge Theory; Princeton University Press: Princeton, NJ, USA, 2018. [Google Scholar]
  39. Strominger, A.; Zhiboedov, A. Gravitational Memory, BMS Supertranslations and Soft Theorems. J. High Energy Phys. 2016, 2016, 86. [Google Scholar] [CrossRef] [Scilit]
  40. Kapec, D.; Perry, M.; Raclariu, A.M.; Strominger, A. Infrared Divergences in QED, Revisited. Phys. Rev. D 2017, 96, 085002. [Google Scholar] [CrossRef] [Scilit]
  41. Danielson, D.L.; Satishchandran, G.; Wald, R.M. Killing horizons decohere quantum superpositions. Phys. Rev. D 2023, 108, 025007. [Google Scholar] [CrossRef] [Scilit]
  42. Prabhu, K.; Satishchandran, G.; Wald, R.M. Infrared finite scattering theory in quantum field theory and quantum gravity. Phys. Rev. D 2022, 106, 066005. [Google Scholar] [CrossRef] [Scilit]
  43. Schwinger, J.S. Brownian Motion of a Quantum Oscillator. J. Math. Phys. 1961, 2, 407–432. [Google Scholar] [CrossRef] [Scilit]
  44. Keldysh, L.V. Diagram Technique for Nonequilibrium Processes. Sov. Phys. JETP 1965, 20, 1018–1026. [Google Scholar]
  45. Feynman, R.P.; Vernon, F.L. The Theory of a General Quantum System Interacting with a Linear Dissipative System. Ann. Phys. 1963, 24, 118–173. [Google Scholar] [CrossRef] [Scilit]
  46. Caldeira, A.O.; Leggett, A.J. Path Integral Approach to Quantum Brownian Motion. Phys. A 1983, 121, 587–616. [Google Scholar] [CrossRef] [Scilit]
  47. Kamenev, A. Field Theory of Non-Equilibrium Systems; Cambridge University Press: Cambridge, UK, 2011. [Google Scholar] [CrossRef] [Scilit]
  48. Cornwall, J.M.; Jackiw, R.; Tomboulis, E. Effective Action for Composite Operators. Phys. Rev. D 1974, 10, 2428–2445. [Google Scholar] [CrossRef] [Scilit]
  49. Okumura, K. Effective Actions of Local Composite Operators: The Case of φ4 Theory, the Itinerant Electron Model, and QED. Int. J. Mod. Phys. A 1996, 11, 65–109. [Google Scholar] [CrossRef] [Scilit]
  50. Aarts, G.; Berges, J. Classical Aspects of Quantum Fields Far from Equilibrium. Phys. Rev. Lett. 2002, 88, 041603. [Google Scholar] [CrossRef] [Scilit] [PubMed]
  51. Hubbard, J. Calculation of Partition Functions. Phys. Rev. Lett. 1959, 3, 77–78. [Google Scholar] [CrossRef] [Scilit]
  52. Stratonovich, R.L. On a Method of Calculating Quantum Distribution Functions. Dokl. Akad. Nauk. SSSR 1957, 115, 1097–1100. [Google Scholar]
  53. Low, F.E. Bremsstrahlung of Very Low-Energy Quanta in Elementary Particle Collisions. Phys. Rev. 1958, 110, 974–977. [Google Scholar] [CrossRef] [Scilit]
  54. Glauber, R.J. The Quantum Theory of Optical Coherence. Phys. Rev. 1963, 130, 2529–2539. [Google Scholar] [CrossRef] [Scilit]
  55. Clerk, A.A.; Devoret, M.H.; Girvin, S.M.; Marquardt, F.; Schoelkopf, R.J. Introduction to Quantum Noise, Measurement, and Amplification. Rev. Mod. Phys. 2010, 82, 1155–1208. [Google Scholar] [CrossRef] [Scilit]
  56. Wald, R.M. Quantum Field Theory in Curved Spacetime and Black Hole Thermodynamics; University of Chicago Press: Chicago, IL, USA, 1994. [Google Scholar]
  57. Décanini, Y.; Folacci, A. Hadamard Renormalization of the Stress-Energy Tensor for a Quantized Scalar Field in a General Spacetime of Arbitrary Dimension. Phys. Rev. D 2008, 78, 044025. [Google Scholar] [CrossRef] [Scilit]
  58. Cutkosky, R.E. Singularities and Discontinuities of Feynman Amplitudes. J. Math. Phys. 1960, 1, 429–433. [Google Scholar] [CrossRef] [Scilit]
  59. Weldon, H.A. Simple Rules for Discontinuities in Finite-Temperature Field Theory. Phys. Rev. D 1983, 28, 2007–2015. [Google Scholar] [CrossRef] [Scilit]
  60. Kobes, R.L.; Semenoff, G.W. Discontinuities of Green Functions in Field Theory at Finite Temperature and Density. Nucl. Phys. B 1985, 260, 714–746. [Google Scholar] [CrossRef] [Scilit]
  61. Kobes, R.L.; Semenoff, G.W. Discontinuities of Green Functions in Field Theory at Finite Temperature and Density. II. Nucl. Phys. B 1986, 272, 329–364. [Google Scholar] [CrossRef] [Scilit]
  62. Landsman, N.P.; van Weert, C.G. Real- and Imaginary-Time Field Theory at Finite Temperature and Density. Phys. Rep. 1987, 145, 141–249. [Google Scholar] [CrossRef] [Scilit]
  63. Ward, J.C. An Identity in Quantum Electrodynamics. Phys. Rev. 1950, 78, 182. [Google Scholar] [CrossRef] [Scilit]
  64. Takahashi, Y. On the Generalized Ward Identity. Il Nuovo Cimento 1957, 6, 371–375. [Google Scholar] [CrossRef] [Scilit]
  65. Hooge, F.N. 1/f Noise Is No Surface Effect. Phys. Lett. A 1969, 29, 139–140. [Google Scholar] [CrossRef] [Scilit]
  66. Johnson, J.B. Thermal Agitation of Electricity in Conductors. Phys. Rev. 1928, 32, 97–109. [Google Scholar] [CrossRef] [Scilit]
  67. Nyquist, H. Thermal Agitation of Electric Charge in Conductors. Phys. Rev. 1928, 32, 110–113. [Google Scholar] [CrossRef] [Scilit]
  68. Callen, H.B.; Welton, T.A. Irreversibility and Generalized Noise. Phys. Rev. 1951, 83, 34–40. [Google Scholar] [CrossRef] [Scilit]
  69. Neuenfeld, D. Infrared-Safe Scattering without Photon Vacuum Transitions and Time-Dependent Decoherence. J. High Energy Phys. 2021, 2021, 189. [Google Scholar] [CrossRef] [Scilit]
  70. Weinberg, S. Infrared Photons and Gravitons. Phys. Rev. 1965, 140, B516–B524. [Google Scholar] [CrossRef] [Scilit]
  71. Christodoulou, D. Nonlinear Nature of Gravitation and Gravitational-Wave Experiments. Phys. Rev. Lett. 1991, 67, 1486–1489. [Google Scholar] [CrossRef] [Scilit] [PubMed]
  72. Favata, M. The Gravitational-Wave Memory Effect. Class. Quantum Gravity 2010, 27, 084036. [Google Scholar] [CrossRef] [Scilit]
  73. Gleiser, M.; Ramos, R.O. Microphysical Approach to Nonequilibrium Dynamics of Quantum Fields. Phys. Rev. D 1994, 50, 2441–2455. [Google Scholar] [CrossRef] [Scilit] [PubMed]
  74. Goldenfeld, N. Lectures on Phase Transitions and the Renormalization Group; Westview Press: Reading, MA, 1992. [Google Scholar]
  75. Chaikin, P.M.; Lubensky, T.C. Principles of Condensed Matter Physics; Cambridge University Press: Cambridge, UK, 1995. [Google Scholar]
  76. Mukhanov, V.F.; Feldman, H.A.; Brandenberger, R.H. Theory of Cosmological Perturbations. Phys. Rep. 1992, 215, 203–333. [Google Scholar] [CrossRef] [Scilit]
  77. Liddle, A.R.; Lyth, D.H. Cosmological Inflation and Large-Scale Structure; Cambridge University Press: Cambridge, UK, 2000. [Google Scholar]
  78. Marcinkiewicz, J. Sur une propriété de la loi de Gauss. Math. Z. 1939, 44, 612–618. [Google Scholar] [CrossRef] [Scilit]
  79. Pattison, C.; Vennin, V.; Wands, D.; Assadullahi, H. Ultra-Slow-Roll Inflation with Quantum Diffusion. J. Cosmol. Astropart. Phys. 2021, 2021, 80. [Google Scholar] [CrossRef] [Scilit]
  80. Hooshangi, S.; Namjoo, M.H.; Noorbala, M. Rare Events Are Nonperturbative: Primordial Black Holes from Heavy-Tailed Distributions. Phys. Lett. B 2022, 834, 137400. [Google Scholar] [CrossRef] [Scilit]
  81. Hsiang, J.T.; Hu, B.L. No Intrinsic Decoherence of Inflationary Cosmological Perturbations. Universe 2022, 8, 27. [Google Scholar] [CrossRef] [Scilit]
  82. Zurek, W.H. Decoherence, einselection, and the quantum origins of the classical. Rev. Mod. Phys. 2003, 75, 715–775. [Google Scholar] [CrossRef] [Scilit]
  83. Morikawa, M.; Nakamichi, A. Quantum Measurement Driven by Spontaneous Symmetry Breaking. Prog. Theor. Phys. 2006, 116, 679–698. [Google Scholar] [CrossRef] [Scilit][Green Version]
  84. Anderson, P.W. Infrared Catastrophe in Fermi Gases with Local Scattering Potentials. Phys. Rev. Lett. 1967, 18, 1049–1051. [Google Scholar] [CrossRef] [Scilit]
  85. Mahan, G.D. Many-Particle Physics, 3rd ed.; Kluwer Academic/Plenum: Dordrecht, The Netherlands, 2000. [Google Scholar]
  86. Nozières, P.; De Dominicis, C.T. Singularities in the X-Ray Absorption and Emission of Metals. III. One-Body Theory Exact Solution. Phys. Rev. 1969, 178, 1097–1107. [Google Scholar] [CrossRef] [Scilit]
  87. Hentschel, M.; Ullmo, D.; Baranger, H.U. Fermi-Edge Singularities in the Mesoscopic X-Ray Edge Problem. Phys. Rev. B 2005, 72, 035310. [Google Scholar] [CrossRef] [Scilit]
  88. Anderson, P.W. The Resonating Valence Bond State in La2CuO4 and Superconductivity. Science 1987, 235, 1196–1198. [Google Scholar] [CrossRef] [Scilit] [PubMed]
  89. Pauling, L.; Wheland, G.W. The Nature of the Chemical Bond. V. The Quantum-Mechanical Calculation of the Resonance Energy of Benzene and Naphthalene and the Hydrocarbon Free Radicals. J. Chem. Phys. 1933, 1, 362–374. [Google Scholar] [CrossRef] [Scilit]
  90. Schafaschek, D.F.; Vasconcelos, G.L.; Macêdo, A.M.S. Tunneling in Double-Well Potentials within Nelson’s Stochastic Mechanics: Application to Ammonia Inversion. Phys. Rev. A 2026, 113, 042214. [Google Scholar] [CrossRef] [Scilit]
  91. Claverie, P.; Jona-Lasinio, G. Instability of Tunneling and the Concept of Molecular Structure in Quantum Mechanics: The Case of Pyramidal Molecules and the Enantiomer Problem. Phys. Rev. A 1986, 33, 2245–2253. [Google Scholar] [CrossRef] [Scilit] [PubMed]
  92. Jona-Lasinio, G.; Presilla, C.; Toninelli, C. Interaction Induced Localization in a Gas of Pyramidal Molecules. Phys. Rev. Lett. 2002, 88, 123001. [Google Scholar] [CrossRef] [Scilit] [PubMed]
  93. Joos, E.; Zeh, H.D. The Emergence of Classical Properties through Interaction with the Environment. Z. Phys. B 1985, 59, 223–243. [Google Scholar] [CrossRef] [Scilit]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Article Metrics

Citations

Article Access Statistics

Multiple requests from the same IP address are counted as one view.